libpostproc/postprocess_template.c

   1 /*
   2  * Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19 */
  20
  21 /**
  22  * @file postprocess_template.c
  23  * mmx/mmx2/3dnow postprocess code.
  24  */
  25
  26
  27 #ifdef ARCH_X86_64
  28 #  define REGa  rax
  29 #  define REGc  rcx
  30 #  define REGd  rdx
  31 #  define REG_a  "rax"
  32 #  define REG_c  "rcx"
  33 #  define REG_d  "rdx"
  34 #  define REG_SP "rsp"
  35 #  define ALIGN_MASK "$0xFFFFFFFFFFFFFFF8"
  36 #else
  37 #  define REGa  eax
  38 #  define REGc  ecx
  39 #  define REGd  edx
  40 #  define REG_a  "eax"
  41 #  define REG_c  "ecx"
  42 #  define REG_d  "edx"
  43 #  define REG_SP "esp"
  44 #  define ALIGN_MASK "$0xFFFFFFF8"
  45 #endif
  46
  47
  48 #undef PAVGB
  49 #undef PMINUB
  50 #undef PMAXUB
  51
  52 #ifdef HAVE_MMX2
  53 #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
  54 #elif defined (HAVE_3DNOW)
  55 #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
  56 #endif
  57 #define PAVGB(a,b)  REAL_PAVGB(a,b)
  58
  59 #ifdef HAVE_MMX2
  60 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
  61 #elif defined (HAVE_MMX)
  62 #define PMINUB(b,a,t) \
  63         "movq " #a ", " #t " \n\t"\
  64         "psubusb " #b ", " #t " \n\t"\
  65         "psubb " #t ", " #a " \n\t"
  66 #endif
  67
  68 #ifdef HAVE_MMX2
  69 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
  70 #elif defined (HAVE_MMX)
  71 #define PMAXUB(a,b) \
  72         "psubusb " #a ", " #b " \n\t"\
  73         "paddb " #a ", " #b " \n\t"
  74 #endif
  75
  76 //FIXME? |255-0| = 1 (shouldnt be a problem ...)
  77 #ifdef HAVE_MMX
  78 /**
  79  * Check if the middle 8x8 Block in the given 8x16 block is flat
  80  */
  81 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
  82         int numEq= 0, dcOk;
  83         src+= stride*4; // src points to begin of the 8x8 Block
  84 asm volatile(
  85                 "movq %0, %%mm7                         \n\t"
  86                 "movq %1, %%mm6                         \n\t"
  87                 : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
  88                 );
  89
  90 asm volatile(
  91                 "lea (%2, %3), %%"REG_a"                \n\t"
  92 //      0       1       2       3       4       5       6       7       8       9
  93 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
  94
  95                 "movq (%2), %%mm0                       \n\t"
  96                 "movq (%%"REG_a"), %%mm1                \n\t"
  97                 "movq %%mm0, %%mm3                      \n\t"
  98                 "movq %%mm0, %%mm4                      \n\t"
  99                 PMAXUB(%%mm1, %%mm4)
 100                 PMINUB(%%mm1, %%mm3, %%mm5)
 101                 "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
 102                 "paddb %%mm7, %%mm0                     \n\t"
 103                 "pcmpgtb %%mm6, %%mm0                   \n\t"
 104
 105                 "movq (%%"REG_a",%3), %%mm2             \n\t"
 106                 PMAXUB(%%mm2, %%mm4)
 107                 PMINUB(%%mm2, %%mm3, %%mm5)
 108                 "psubb %%mm2, %%mm1                     \n\t"
 109                 "paddb %%mm7, %%mm1                     \n\t"
 110                 "pcmpgtb %%mm6, %%mm1                   \n\t"
 111                 "paddb %%mm1, %%mm0                     \n\t"
 112
 113                 "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
 114                 PMAXUB(%%mm1, %%mm4)
 115                 PMINUB(%%mm1, %%mm3, %%mm5)
 116                 "psubb %%mm1, %%mm2                     \n\t"
 117                 "paddb %%mm7, %%mm2                     \n\t"
 118                 "pcmpgtb %%mm6, %%mm2                   \n\t"
 119                 "paddb %%mm2, %%mm0                     \n\t"
 120
 121                 "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
 122
 123                 "movq (%2, %3, 4), %%mm2                \n\t"
 124                 PMAXUB(%%mm2, %%mm4)
 125                 PMINUB(%%mm2, %%mm3, %%mm5)
 126                 "psubb %%mm2, %%mm1                     \n\t"
 127                 "paddb %%mm7, %%mm1                     \n\t"
 128                 "pcmpgtb %%mm6, %%mm1                   \n\t"
 129                 "paddb %%mm1, %%mm0                     \n\t"
 130
 131                 "movq (%%"REG_a"), %%mm1                \n\t"
 132                 PMAXUB(%%mm1, %%mm4)
 133                 PMINUB(%%mm1, %%mm3, %%mm5)
 134                 "psubb %%mm1, %%mm2                     \n\t"
 135                 "paddb %%mm7, %%mm2                     \n\t"
 136                 "pcmpgtb %%mm6, %%mm2                   \n\t"
 137                 "paddb %%mm2, %%mm0                     \n\t"
 138
 139                 "movq (%%"REG_a", %3), %%mm2            \n\t"
 140                 PMAXUB(%%mm2, %%mm4)
 141                 PMINUB(%%mm2, %%mm3, %%mm5)
 142                 "psubb %%mm2, %%mm1                     \n\t"
 143                 "paddb %%mm7, %%mm1                     \n\t"
 144                 "pcmpgtb %%mm6, %%mm1                   \n\t"
 145                 "paddb %%mm1, %%mm0                     \n\t"
 146
 147                 "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
 148                 PMAXUB(%%mm1, %%mm4)
 149                 PMINUB(%%mm1, %%mm3, %%mm5)
 150                 "psubb %%mm1, %%mm2                     \n\t"
 151                 "paddb %%mm7, %%mm2                     \n\t"
 152                 "pcmpgtb %%mm6, %%mm2                   \n\t"
 153                 "paddb %%mm2, %%mm0                     \n\t"
 154                 "psubusb %%mm3, %%mm4                   \n\t"
 155
 156                 "                                       \n\t"
 157 #ifdef HAVE_MMX2
 158                 "pxor %%mm7, %%mm7                      \n\t"
 159                 "psadbw %%mm7, %%mm0                    \n\t"
 160 #else
 161                 "movq %%mm0, %%mm1                      \n\t"
 162                 "psrlw $8, %%mm0                        \n\t"
 163                 "paddb %%mm1, %%mm0                     \n\t"
 164                 "movq %%mm0, %%mm1                      \n\t"
 165                 "psrlq $16, %%mm0                       \n\t"
 166                 "paddb %%mm1, %%mm0                     \n\t"
 167                 "movq %%mm0, %%mm1                      \n\t"
 168                 "psrlq $32, %%mm0                       \n\t"
 169                 "paddb %%mm1, %%mm0                     \n\t"
 170 #endif
 171                 "movq %4, %%mm7                         \n\t" // QP,..., QP
 172                 "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
 173                 "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
 174                 "packssdw %%mm4, %%mm4                  \n\t"
 175                 "movd %%mm0, %0                         \n\t"
 176                 "movd %%mm4, %1                         \n\t"
 177
 178                 : "=r" (numEq), "=r" (dcOk)
 179                 : "r" (src), "r" ((long)stride), "m" (c->pQPb)
 180                 : "%"REG_a
 181                 );
 182
 183         numEq= (-numEq) &0xFF;
 184         if(numEq > c->ppMode.flatnessThreshold){
 185             if(dcOk) return 0;
 186             else     return 1;
 187         }else{
 188             return 2;
 189         }
 190 }
 191 #endif //HAVE_MMX
 192
 193 /**
 194  * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
 195  * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
 196  */
 197 #ifndef HAVE_ALTIVEC
 198 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
 199 {
 200 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 201         src+= stride*3;
 202         asm volatile(        //"movv %0 %1 %2\n\t"
 203                 "movq %2, %%mm0                         \n\t"  // QP,..., QP
 204                 "pxor %%mm4, %%mm4                      \n\t"
 205
 206                 "movq (%0), %%mm6                       \n\t"
 207                 "movq (%0, %1), %%mm5                   \n\t"
 208                 "movq %%mm5, %%mm1                      \n\t"
 209                 "movq %%mm6, %%mm2                      \n\t"
 210                 "psubusb %%mm6, %%mm5                   \n\t"
 211                 "psubusb %%mm1, %%mm2                   \n\t"
 212                 "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
 213                 "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
 214                 "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
 215
 216                 "pand %%mm2, %%mm6                      \n\t"
 217                 "pandn %%mm1, %%mm2                     \n\t"
 218                 "por %%mm2, %%mm6                       \n\t"// First Line to Filter
 219
 220                 "movq (%0, %1, 8), %%mm5                \n\t"
 221                 "lea (%0, %1, 4), %%"REG_a"             \n\t"
 222                 "lea (%0, %1, 8), %%"REG_c"             \n\t"
 223                 "sub %1, %%"REG_c"                      \n\t"
 224                 "add %1, %0                             \n\t" // %0 points to line 1 not 0
 225                 "movq (%0, %1, 8), %%mm7                \n\t"
 226                 "movq %%mm5, %%mm1                      \n\t"
 227                 "movq %%mm7, %%mm2                      \n\t"
 228                 "psubusb %%mm7, %%mm5                   \n\t"
 229                 "psubusb %%mm1, %%mm2                   \n\t"
 230                 "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
 231                 "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
 232                 "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
 233
 234                 "pand %%mm2, %%mm7                      \n\t"
 235                 "pandn %%mm1, %%mm2                     \n\t"
 236                 "por %%mm2, %%mm7                       \n\t" // First Line to Filter
 237
 238
 239                 //      1       2       3       4       5       6       7       8
 240                 //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
 241                 // 6 4 2 2 1 1
 242                 // 6 4 4 2
 243                 // 6 8 2
 244
 245                 "movq (%0, %1), %%mm0                   \n\t" //  1
 246                 "movq %%mm0, %%mm1                      \n\t" //  1
 247                 PAVGB(%%mm6, %%mm0)                           //1 1        /2
 248                 PAVGB(%%mm6, %%mm0)                           //3 1        /4
 249
 250                 "movq (%0, %1, 4), %%mm2                \n\t" //     1
 251                 "movq %%mm2, %%mm5                      \n\t" //     1
 252                 PAVGB((%%REGa), %%mm2)                        //    11        /2
 253                 PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
 254                 "movq %%mm2, %%mm3                      \n\t" //   211        /4
 255                 "movq (%0), %%mm4                       \n\t" // 1
 256                 PAVGB(%%mm4, %%mm3)                           // 4 211        /8
 257                 PAVGB(%%mm0, %%mm3)                           //642211        /16
 258                 "movq %%mm3, (%0)                       \n\t" // X
 259                 // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
 260                 "movq %%mm1, %%mm0                      \n\t" //  1
 261                 PAVGB(%%mm6, %%mm0)                           //1 1        /2
 262                 "movq %%mm4, %%mm3                      \n\t" // 1
 263                 PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
 264                 PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
 265                 PAVGB((%%REGa), %%mm5)                        //    211 /4
 266                 PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
 267                 PAVGB(%%mm0, %%mm3)                           //4242211 /16
 268                 "movq %%mm3, (%0,%1)                    \n\t" //  X
 269                 // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
 270                 PAVGB(%%mm4, %%mm6)                                   //11        /2
 271                 "movq (%%"REG_c"), %%mm0                \n\t" //       1
 272                 PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
 273                 "movq %%mm0, %%mm3                      \n\t" //      11/2
 274                 PAVGB(%%mm1, %%mm0)                           //  2   11/4
 275                 PAVGB(%%mm6, %%mm0)                           //222   11/8
 276                 PAVGB(%%mm2, %%mm0)                           //22242211/16
 277                 "movq (%0, %1, 2), %%mm2                \n\t" //   1
 278                 "movq %%mm0, (%0, %1, 2)                \n\t" //   X
 279                 // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
 280                 "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
 281                 PAVGB((%%REGc), %%mm0)                        //       11        /2
 282                 PAVGB(%%mm0, %%mm6)                           //11     11        /4
 283                 PAVGB(%%mm1, %%mm4)                           // 11                /2
 284                 PAVGB(%%mm2, %%mm1)                           //  11                /2
 285                 PAVGB(%%mm1, %%mm6)                           //1122   11        /8
 286                 PAVGB(%%mm5, %%mm6)                           //112242211        /16
 287                 "movq (%%"REG_a"), %%mm5                \n\t" //    1
 288                 "movq %%mm6, (%%"REG_a")                \n\t" //    X
 289                 // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
 290                 "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
 291                 PAVGB(%%mm7, %%mm6)                           //        11        /2
 292                 PAVGB(%%mm4, %%mm6)                           // 11     11        /4
 293                 PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
 294                 PAVGB(%%mm5, %%mm2)                           //   11                /2
 295                 "movq (%0, %1, 4), %%mm4                \n\t" //     1
 296                 PAVGB(%%mm4, %%mm2)                           //   112                /4
 297                 PAVGB(%%mm2, %%mm6)                           // 112242211        /16
 298                 "movq %%mm6, (%0, %1, 4)                \n\t" //     X
 299                 // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
 300                 PAVGB(%%mm7, %%mm1)                           //  11     2        /4
 301                 PAVGB(%%mm4, %%mm5)                           //    11                /2
 302                 PAVGB(%%mm5, %%mm0)                           //    11 11        /4
 303                 "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
 304                 PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
 305                 PAVGB(%%mm0, %%mm1)                           //  11224222        /16
 306                 "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
 307                 // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
 308                 PAVGB((%%REGc), %%mm2)                        //   112 4        /8
 309                 "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
 310                 PAVGB(%%mm0, %%mm6)                           //      1 1        /2
 311                 PAVGB(%%mm7, %%mm6)                           //      1 12        /4
 312                 PAVGB(%%mm2, %%mm6)                           //   1122424        /4
 313                 "movq %%mm6, (%%"REG_c")                \n\t" //       X
 314                 // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
 315                 PAVGB(%%mm7, %%mm5)                           //    11   2        /4
 316                 PAVGB(%%mm7, %%mm5)                           //    11   6        /8
 317
 318                 PAVGB(%%mm3, %%mm0)                           //      112        /4
 319                 PAVGB(%%mm0, %%mm5)                           //    112246        /16
 320                 "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
 321                 "sub %1, %0                             \n\t"
 322
 323                 :
 324                 : "r" (src), "r" ((long)stride), "m" (c->pQPb)
 325                 : "%"REG_a, "%"REG_c
 326         );
 327 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 328         const int l1= stride;
 329         const int l2= stride + l1;
 330         const int l3= stride + l2;
 331         const int l4= stride + l3;
 332         const int l5= stride + l4;
 333         const int l6= stride + l5;
 334         const int l7= stride + l6;
 335         const int l8= stride + l7;
 336         const int l9= stride + l8;
 337         int x;
 338         src+= stride*3;
 339         for(x=0; x<BLOCK_SIZE; x++)
 340         {
 341                 const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
 342                 const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
 343
 344                 int sums[10];
 345                 sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
 346                 sums[1] = sums[0] - first  + src[l4];
 347                 sums[2] = sums[1] - first  + src[l5];
 348                 sums[3] = sums[2] - first  + src[l6];
 349                 sums[4] = sums[3] - first  + src[l7];
 350                 sums[5] = sums[4] - src[l1] + src[l8];
 351                 sums[6] = sums[5] - src[l2] + last;
 352                 sums[7] = sums[6] - src[l3] + last;
 353                 sums[8] = sums[7] - src[l4] + last;
 354                 sums[9] = sums[8] - src[l5] + last;
 355
 356                 src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
 357                 src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
 358                 src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
 359                 src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
 360                 src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
 361                 src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
 362                 src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
 363                 src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
 364
 365                 src++;
 366         }
 367 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 368 }
 369 #endif //HAVE_ALTIVEC
 370
 371 #if 0
 372 /**
 373  * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
 374  * values are correctly clipped (MMX2)
 375  * values are wraparound (C)
 376  * conclusion: its fast, but introduces ugly horizontal patterns if there is a continious gradient
 377         0 8 16 24
 378         x = 8
 379         x/2 = 4
 380         x/8 = 1
 381         1 12 12 23
 382  */
 383 static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
 384 {
 385 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 386         src+= stride*3;
 387 // FIXME rounding
 388         asm volatile(
 389                 "pxor %%mm7, %%mm7                      \n\t" // 0
 390                 "movq "MANGLE(b80)", %%mm6              \n\t" // MIN_SIGNED_BYTE
 391                 "leal (%0, %1), %%"REG_a"               \n\t"
 392                 "leal (%%"REG_a", %1, 4), %%"REG_c"     \n\t"
 393 //      0       1       2       3       4       5       6       7       8       9
 394 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
 395                 "movq "MANGLE(pQPb)", %%mm0             \n\t" // QP,..., QP
 396                 "movq %%mm0, %%mm1                      \n\t" // QP,..., QP
 397                 "paddusb "MANGLE(b02)", %%mm0           \n\t"
 398                 "psrlw $2, %%mm0                        \n\t"
 399                 "pand "MANGLE(b3F)", %%mm0              \n\t" // QP/4,..., QP/4
 400                 "paddusb %%mm1, %%mm0                   \n\t" // QP*1.25 ...
 401                 "movq (%0, %1, 4), %%mm2                \n\t" // line 4
 402                 "movq (%%"REG_c"), %%mm3                \n\t" // line 5
 403                 "movq %%mm2, %%mm4                      \n\t" // line 4
 404                 "pcmpeqb %%mm5, %%mm5                   \n\t" // -1
 405                 "pxor %%mm2, %%mm5                      \n\t" // -line 4 - 1
 406                 PAVGB(%%mm3, %%mm5)
 407                 "paddb %%mm6, %%mm5                     \n\t" // (l5-l4)/2
 408                 "psubusb %%mm3, %%mm4                   \n\t"
 409                 "psubusb %%mm2, %%mm3                   \n\t"
 410                 "por %%mm3, %%mm4                       \n\t" // |l4 - l5|
 411                 "psubusb %%mm0, %%mm4                   \n\t"
 412                 "pcmpeqb %%mm7, %%mm4                   \n\t"
 413                 "pand %%mm4, %%mm5                      \n\t" // d/2
 414
 415 //                "paddb %%mm6, %%mm2                     \n\t" // line 4 + 0x80
 416                 "paddb %%mm5, %%mm2                     \n\t"
 417 //                "psubb %%mm6, %%mm2                     \n\t"
 418                 "movq %%mm2, (%0,%1, 4)                 \n\t"
 419
 420                 "movq (%%"REG_c"), %%mm2                \n\t"
 421 //                "paddb %%mm6, %%mm2                     \n\t" // line 5 + 0x80
 422                 "psubb %%mm5, %%mm2                     \n\t"
 423 //                "psubb %%mm6, %%mm2                     \n\t"
 424                 "movq %%mm2, (%%"REG_c")                \n\t"
 425
 426                 "paddb %%mm6, %%mm5                     \n\t"
 427                 "psrlw $2, %%mm5                        \n\t"
 428                 "pand "MANGLE(b3F)", %%mm5              \n\t"
 429                 "psubb "MANGLE(b20)", %%mm5             \n\t" // (l5-l4)/8
 430
 431                 "movq (%%"REG_a", %1, 2), %%mm2         \n\t"
 432                 "paddb %%mm6, %%mm2                     \n\t" // line 3 + 0x80
 433                 "paddsb %%mm5, %%mm2                    \n\t"
 434                 "psubb %%mm6, %%mm2                     \n\t"
 435                 "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
 436
 437                 "movq (%%"REG_c", %1), %%mm2            \n\t"
 438                 "paddb %%mm6, %%mm2                     \n\t" // line 6 + 0x80
 439                 "psubsb %%mm5, %%mm2                    \n\t"
 440                 "psubb %%mm6, %%mm2                     \n\t"
 441                 "movq %%mm2, (%%"REG_c", %1)            \n\t"
 442
 443                 :
 444                 : "r" (src), "r" ((long)stride)
 445                 : "%"REG_a, "%"REG_c
 446         );
 447 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 448          const int l1= stride;
 449         const int l2= stride + l1;
 450         const int l3= stride + l2;
 451         const int l4= stride + l3;
 452         const int l5= stride + l4;
 453         const int l6= stride + l5;
 454 //        const int l7= stride + l6;
 455 //        const int l8= stride + l7;
 456 //        const int l9= stride + l8;
 457         int x;
 458         const int QP15= QP + (QP>>2);
 459         src+= stride*3;
 460         for(x=0; x<BLOCK_SIZE; x++)
 461         {
 462                 const int v = (src[x+l5] - src[x+l4]);
 463                 if(FFABS(v) < QP15)
 464                 {
 465                         src[x+l3] +=v>>3;
 466                         src[x+l4] +=v>>1;
 467                         src[x+l5] -=v>>1;
 468                         src[x+l6] -=v>>3;
 469
 470                 }
 471         }
 472
 473 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 474 }
 475 #endif //0
 476
 477 /**
 478  * Experimental Filter 1
 479  * will not damage linear gradients
 480  * Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
 481  * can only smooth blocks at the expected locations (it cant smooth them if they did move)
 482  * MMX2 version does correct clipping C version doesnt
 483  */
 484 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
 485 {
 486 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 487         src+= stride*3;
 488
 489         asm volatile(
 490                 "pxor %%mm7, %%mm7                      \n\t" // 0
 491                 "lea (%0, %1), %%"REG_a"                \n\t"
 492                 "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
 493 //      0       1       2       3       4       5       6       7       8       9
 494 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
 495                 "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
 496                 "movq (%0, %1, 4), %%mm1                \n\t" // line 4
 497                 "movq %%mm1, %%mm2                      \n\t" // line 4
 498                 "psubusb %%mm0, %%mm1                   \n\t"
 499                 "psubusb %%mm2, %%mm0                   \n\t"
 500                 "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
 501                 "movq (%%"REG_c"), %%mm3                \n\t" // line 5
 502                 "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
 503                 "movq %%mm3, %%mm5                      \n\t" // line 5
 504                 "psubusb %%mm4, %%mm3                   \n\t"
 505                 "psubusb %%mm5, %%mm4                   \n\t"
 506                 "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
 507                 PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
 508                 "movq %%mm2, %%mm1                      \n\t" // line 4
 509                 "psubusb %%mm5, %%mm2                   \n\t"
 510                 "movq %%mm2, %%mm4                      \n\t"
 511                 "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
 512                 "psubusb %%mm1, %%mm5                   \n\t"
 513                 "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
 514                 "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
 515                 "movq %%mm4, %%mm3                      \n\t" // d
 516                 "movq %2, %%mm0                         \n\t"
 517                 "paddusb %%mm0, %%mm0                   \n\t"
 518                 "psubusb %%mm0, %%mm4                   \n\t"
 519                 "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
 520                 "psubusb "MANGLE(b01)", %%mm3           \n\t"
 521                 "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
 522
 523                 PAVGB(%%mm7, %%mm3)                           // d/2
 524                 "movq %%mm3, %%mm1                      \n\t" // d/2
 525                 PAVGB(%%mm7, %%mm3)                           // d/4
 526                 PAVGB(%%mm1, %%mm3)                           // 3*d/8
 527
 528                 "movq (%0, %1, 4), %%mm0                \n\t" // line 4
 529                 "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
 530                 "psubusb %%mm3, %%mm0                   \n\t"
 531                 "pxor %%mm2, %%mm0                      \n\t"
 532                 "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
 533
 534                 "movq (%%"REG_c"), %%mm0                \n\t" // line 5
 535                 "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
 536                 "paddusb %%mm3, %%mm0                   \n\t"
 537                 "pxor %%mm2, %%mm0                      \n\t"
 538                 "movq %%mm0, (%%"REG_c")                \n\t" // line 5
 539
 540                 PAVGB(%%mm7, %%mm1)                           // d/4
 541
 542                 "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
 543                 "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
 544                 "psubusb %%mm1, %%mm0                   \n\t"
 545                 "pxor %%mm2, %%mm0                      \n\t"
 546                 "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
 547
 548                 "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
 549                 "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
 550                 "paddusb %%mm1, %%mm0                   \n\t"
 551                 "pxor %%mm2, %%mm0                      \n\t"
 552                 "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
 553
 554                 PAVGB(%%mm7, %%mm1)                           // d/8
 555
 556                 "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
 557                 "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
 558                 "psubusb %%mm1, %%mm0                   \n\t"
 559                 "pxor %%mm2, %%mm0                      \n\t"
 560                 "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
 561
 562                 "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
 563                 "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
 564                 "paddusb %%mm1, %%mm0                   \n\t"
 565                 "pxor %%mm2, %%mm0                      \n\t"
 566                 "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
 567
 568                 :
 569                 : "r" (src), "r" ((long)stride), "m" (co->pQPb)
 570                 : "%"REG_a, "%"REG_c
 571         );
 572 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 573
 574         const int l1= stride;
 575         const int l2= stride + l1;
 576         const int l3= stride + l2;
 577         const int l4= stride + l3;
 578         const int l5= stride + l4;
 579         const int l6= stride + l5;
 580         const int l7= stride + l6;
 581 //        const int l8= stride + l7;
 582 //        const int l9= stride + l8;
 583         int x;
 584
 585         src+= stride*3;
 586         for(x=0; x<BLOCK_SIZE; x++)
 587         {
 588                 int a= src[l3] - src[l4];
 589                 int b= src[l4] - src[l5];
 590                 int c= src[l5] - src[l6];
 591
 592                 int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
 593                 d= FFMAX(d, 0);
 594
 595                 if(d < co->QP*2)
 596                 {
 597                         int v = d * FFSIGN(-b);
 598
 599                         src[l2] +=v>>3;
 600                         src[l3] +=v>>2;
 601                         src[l4] +=(3*v)>>3;
 602                         src[l5] -=(3*v)>>3;
 603                         src[l6] -=v>>2;
 604                         src[l7] -=v>>3;
 605
 606                 }
 607                 src++;
 608         }
 609 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 610 }
 611
 612 #ifndef HAVE_ALTIVEC
 613 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
 614 {
 615 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 616 /*
 617         uint8_t tmp[16];
 618         const int l1= stride;
 619         const int l2= stride + l1;
 620         const int l3= stride + l2;
 621         const int l4= (int)tmp - (int)src - stride*3;
 622         const int l5= (int)tmp - (int)src - stride*3 + 8;
 623         const int l6= stride*3 + l3;
 624         const int l7= stride + l6;
 625         const int l8= stride + l7;
 626
 627         memcpy(tmp, src+stride*7, 8);
 628         memcpy(tmp+8, src+stride*8, 8);
 629 */
 630         src+= stride*4;
 631         asm volatile(
 632
 633 #if 0 //sligtly more accurate and slightly slower
 634                 "pxor %%mm7, %%mm7                      \n\t" // 0
 635                 "lea (%0, %1), %%"REG_a"                \n\t"
 636                 "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
 637 //      0       1       2       3       4       5       6       7
 638 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
 639 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
 640
 641
 642                 "movq (%0, %1, 2), %%mm0                \n\t" // l2
 643                 "movq (%0), %%mm1                       \n\t" // l0
 644                 "movq %%mm0, %%mm2                      \n\t" // l2
 645                 PAVGB(%%mm7, %%mm0)                           // ~l2/2
 646                 PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
 647                 PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
 648
 649                 "movq (%%"REG_a"), %%mm1                \n\t" // l1
 650                 "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
 651                 "movq %%mm1, %%mm4                      \n\t" // l1
 652                 PAVGB(%%mm7, %%mm1)                           // ~l1/2
 653                 PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
 654                 PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
 655
 656                 "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
 657                 "psubusb %%mm1, %%mm0                   \n\t"
 658                 "psubusb %%mm4, %%mm1                   \n\t"
 659                 "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
 660 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
 661
 662                 "movq (%0, %1, 4), %%mm0                \n\t" // l4
 663                 "movq %%mm0, %%mm4                      \n\t" // l4
 664                 PAVGB(%%mm7, %%mm0)                           // ~l4/2
 665                 PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
 666                 PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
 667
 668                 "movq (%%"REG_c"), %%mm2                \n\t" // l5
 669                 "movq %%mm3, %%mm5                      \n\t" // l3
 670                 PAVGB(%%mm7, %%mm3)                           // ~l3/2
 671                 PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
 672                 PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
 673
 674                 "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
 675                 "psubusb %%mm3, %%mm0                   \n\t"
 676                 "psubusb %%mm6, %%mm3                   \n\t"
 677                 "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
 678                 "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
 679 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
 680
 681                 "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
 682                 "movq %%mm6, %%mm5                      \n\t" // l6
 683                 PAVGB(%%mm7, %%mm6)                           // ~l6/2
 684                 PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
 685                 PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
 686
 687                 "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
 688                 "movq %%mm2, %%mm4                      \n\t" // l5
 689                 PAVGB(%%mm7, %%mm2)                           // ~l5/2
 690                 PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
 691                 PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
 692
 693                 "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
 694                 "psubusb %%mm2, %%mm6                   \n\t"
 695                 "psubusb %%mm4, %%mm2                   \n\t"
 696                 "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
 697 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
 698
 699
 700                 PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
 701                 "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
 702                 "paddusb "MANGLE(b01)", %%mm4           \n\t"
 703                 "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
 704                 "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
 705                 "pand %%mm4, %%mm3                      \n\t"
 706
 707                 "movq %%mm3, %%mm1                      \n\t"
 708 //                "psubusb "MANGLE(b01)", %%mm3           \n\t"
 709                 PAVGB(%%mm7, %%mm3)
 710                 PAVGB(%%mm7, %%mm3)
 711                 "paddusb %%mm1, %%mm3                   \n\t"
 712 //                "paddusb "MANGLE(b01)", %%mm3           \n\t"
 713
 714                 "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
 715                 "movq (%0, %1, 4), %%mm5                \n\t" //l4
 716                 "movq (%0, %1, 4), %%mm4                \n\t" //l4
 717                 "psubusb %%mm6, %%mm5                   \n\t"
 718                 "psubusb %%mm4, %%mm6                   \n\t"
 719                 "por %%mm6, %%mm5                       \n\t" // |l3-l4|
 720                 "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
 721                 "pxor %%mm6, %%mm0                      \n\t"
 722                 "pand %%mm0, %%mm3                      \n\t"
 723                 PMINUB(%%mm5, %%mm3, %%mm0)
 724
 725                 "psubusb "MANGLE(b01)", %%mm3           \n\t"
 726                 PAVGB(%%mm7, %%mm3)
 727
 728                 "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
 729                 "movq (%0, %1, 4), %%mm2                \n\t"
 730                 "pxor %%mm6, %%mm0                      \n\t"
 731                 "pxor %%mm6, %%mm2                      \n\t"
 732                 "psubb %%mm3, %%mm0                     \n\t"
 733                 "paddb %%mm3, %%mm2                     \n\t"
 734                 "pxor %%mm6, %%mm0                      \n\t"
 735                 "pxor %%mm6, %%mm2                      \n\t"
 736                 "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
 737                 "movq %%mm2, (%0, %1, 4)                \n\t"
 738 #endif //0
 739
 740                 "lea (%0, %1), %%"REG_a"                \n\t"
 741                 "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
 742 //      0       1       2       3       4       5       6       7
 743 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
 744 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
 745
 746
 747                 "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
 748                 "movq (%0, %1, 4), %%mm0                \n\t" // l4
 749                 "pxor %%mm6, %%mm1                      \n\t" // -l3-1
 750                 PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
 751 // mm1=-l3-1, mm0=128-q
 752
 753                 "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
 754                 "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
 755                 "pxor %%mm6, %%mm2                      \n\t" // -l5-1
 756                 "movq %%mm2, %%mm5                      \n\t" // -l5-1
 757                 "movq "MANGLE(b80)", %%mm4              \n\t" // 128
 758                 "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
 759                 PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
 760                 PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
 761                 PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
 762                 PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
 763 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
 764
 765                 "movq (%%"REG_a"), %%mm2                \n\t" // l1
 766                 "pxor %%mm6, %%mm2                      \n\t" // -l1-1
 767                 PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
 768                 PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
 769                 "movq "MANGLE(b80)", %%mm3              \n\t" // 128
 770                 PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
 771                 PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
 772                 PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
 773 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
 774
 775                 PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
 776                 "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
 777                 "pxor %%mm6, %%mm1                      \n\t" // -l7-1
 778                 PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
 779                 "movq "MANGLE(b80)", %%mm2              \n\t" // 128
 780                 PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
 781                 PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
 782                 PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
 783 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
 784
 785                 "movq "MANGLE(b00)", %%mm1              \n\t" // 0
 786                 "movq "MANGLE(b00)", %%mm5              \n\t" // 0
 787                 "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
 788                 "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
 789                 PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
 790                  PMAXUB(%%mm5, %%mm3)                         // 128 + |lenergy/16|
 791                 PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
 792
 793 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
 794
 795                 "movq "MANGLE(b00)", %%mm7              \n\t" // 0
 796                 "movq %2, %%mm2                         \n\t" // QP
 797                 PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
 798                 "psubb %%mm6, %%mm2                     \n\t"
 799
 800                 "movq %%mm4, %%mm1                      \n\t"
 801                 "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
 802                 "pxor %%mm1, %%mm4                      \n\t"
 803                 "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
 804                 "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
 805                 "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
 806 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
 807
 808                 "movq %%mm4, %%mm3                      \n\t" // d
 809                 "psubusb "MANGLE(b01)", %%mm4           \n\t"
 810                 PAVGB(%%mm7, %%mm4)                           // d/32
 811                 PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
 812                 "paddb %%mm3, %%mm4                     \n\t" // 5d/64
 813                 "pand %%mm2, %%mm4                      \n\t"
 814
 815                 "movq "MANGLE(b80)", %%mm5              \n\t" // 128
 816                 "psubb %%mm0, %%mm5                     \n\t" // q
 817                 "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
 818                 "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
 819                 "pxor %%mm7, %%mm5                      \n\t"
 820
 821                 PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
 822                 "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
 823
 824                 "pand %%mm7, %%mm4                      \n\t"
 825                 "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
 826                 "movq (%0, %1, 4), %%mm2                \n\t"
 827                 "pxor %%mm1, %%mm0                      \n\t"
 828                 "pxor %%mm1, %%mm2                      \n\t"
 829                 "paddb %%mm4, %%mm0                     \n\t"
 830                 "psubb %%mm4, %%mm2                     \n\t"
 831                 "pxor %%mm1, %%mm0                      \n\t"
 832                 "pxor %%mm1, %%mm2                      \n\t"
 833                 "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
 834                 "movq %%mm2, (%0, %1, 4)                \n\t"
 835
 836                 :
 837                 : "r" (src), "r" ((long)stride), "m" (c->pQPb)
 838                 : "%"REG_a, "%"REG_c
 839         );
 840
 841 /*
 842         {
 843         int x;
 844         src-= stride;
 845         for(x=0; x<BLOCK_SIZE; x++)
 846         {
 847                 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
 848                 if(FFABS(middleEnergy)< 8*QP)
 849                 {
 850                         const int q=(src[l4] - src[l5])/2;
 851                         const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
 852                         const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
 853
 854                         int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
 855                         d= FFMAX(d, 0);
 856
 857                         d= (5*d + 32) >> 6;
 858                         d*= FFSIGN(-middleEnergy);
 859
 860                         if(q>0)
 861                         {
 862                                 d= d<0 ? 0 : d;
 863                                 d= d>q ? q : d;
 864                         }
 865                         else
 866                         {
 867                                 d= d>0 ? 0 : d;
 868                                 d= d<q ? q : d;
 869                         }
 870
 871                         src[l4]-= d;
 872                         src[l5]+= d;
 873                 }
 874                 src++;
 875         }
 876 src-=8;
 877         for(x=0; x<8; x++)
 878         {
 879                 int y;
 880                 for(y=4; y<6; y++)
 881                 {
 882                         int d= src[x+y*stride] - tmp[x+(y-4)*8];
 883                         int ad= FFABS(d);
 884                         static int max=0;
 885                         static int sum=0;
 886                         static int num=0;
 887                         static int bias=0;
 888
 889                         if(max<ad) max=ad;
 890                         sum+= ad>3 ? 1 : 0;
 891                         if(ad>3)
 892                         {
 893                                 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
 894                         }
 895                         if(y==4) bias+=d;
 896                         num++;
 897                         if(num%1000000 == 0)
 898                         {
 899                                 av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
 900                         }
 901                 }
 902         }
 903 }
 904 */
 905 #elif defined (HAVE_MMX)
 906         src+= stride*4;
 907         asm volatile(
 908                 "pxor %%mm7, %%mm7                      \n\t"
 909                 "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
 910                 "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
 911 //      0       1       2       3       4       5       6       7
 912 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
 913 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1
 914
 915                 "movq (%0), %%mm0                       \n\t"
 916                 "movq %%mm0, %%mm1                      \n\t"
 917                 "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
 918                 "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
 919
 920                 "movq (%0, %1), %%mm2                   \n\t"
 921                 "lea (%0, %1, 2), %%"REG_a"             \n\t"
 922                 "movq %%mm2, %%mm3                      \n\t"
 923                 "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
 924                 "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
 925
 926                 "movq (%%"REG_a"), %%mm4                \n\t"
 927                 "movq %%mm4, %%mm5                      \n\t"
 928                 "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
 929                 "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
 930
 931                 "paddw %%mm0, %%mm0                     \n\t" // 2L0
 932                 "paddw %%mm1, %%mm1                     \n\t" // 2H0
 933                 "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
 934                 "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
 935                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
 936                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
 937
 938                 "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
 939                 "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
 940                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
 941                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
 942
 943                 "movq (%%"REG_a", %1), %%mm2            \n\t"
 944                 "movq %%mm2, %%mm3                      \n\t"
 945                 "punpcklbw %%mm7, %%mm2                 \n\t" // L3
 946                 "punpckhbw %%mm7, %%mm3                 \n\t" // H3
 947
 948                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
 949                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
 950                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
 951                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
 952                 "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
 953                 "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
 954
 955                 "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
 956                 "movq %%mm0, %%mm1                      \n\t"
 957                 "punpcklbw %%mm7, %%mm0                 \n\t" // L4
 958                 "punpckhbw %%mm7, %%mm1                 \n\t" // H4
 959
 960                 "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
 961                 "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
 962                 "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
 963                 "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
 964                 "paddw %%mm4, %%mm4                     \n\t" // 2L2
 965                 "paddw %%mm5, %%mm5                     \n\t" // 2H2
 966                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
 967                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
 968
 969                 "lea (%%"REG_a", %1), %0                \n\t"
 970                 "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
 971                 "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
 972                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
 973                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
 974 //50 opcodes so far
 975                 "movq (%0, %1, 2), %%mm2                \n\t"
 976                 "movq %%mm2, %%mm3                      \n\t"
 977                 "punpcklbw %%mm7, %%mm2                 \n\t" // L5
 978                 "punpckhbw %%mm7, %%mm3                 \n\t" // H5
 979                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
 980                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
 981                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
 982                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
 983
 984                 "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
 985                 "punpcklbw %%mm7, %%mm6                 \n\t" // L6
 986                 "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
 987                 "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
 988                 "punpckhbw %%mm7, %%mm6                 \n\t" // H6
 989                 "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
 990
 991                 "paddw %%mm0, %%mm0                     \n\t" // 2L4
 992                 "paddw %%mm1, %%mm1                     \n\t" // 2H4
 993                 "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
 994                 "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
 995
 996                 "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
 997                 "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
 998                 "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
 999                 "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
1000
1001                 "movq (%0, %1, 4), %%mm2                \n\t"
1002                 "movq %%mm2, %%mm3                      \n\t"
1003                 "punpcklbw %%mm7, %%mm2                 \n\t" // L7
1004                 "punpckhbw %%mm7, %%mm3                 \n\t" // H7
1005
1006                 "paddw %%mm2, %%mm2                     \n\t" // 2L7
1007                 "paddw %%mm3, %%mm3                     \n\t" // 2H7
1008                 "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
1009                 "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
1010
1011                 "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
1012                 "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
1013
1014 #ifdef HAVE_MMX2
1015                 "movq %%mm7, %%mm6                      \n\t" // 0
1016                 "psubw %%mm0, %%mm6                     \n\t"
1017                 "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1018                 "movq %%mm7, %%mm6                      \n\t" // 0
1019                 "psubw %%mm1, %%mm6                     \n\t"
1020                 "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1021                 "movq %%mm7, %%mm6                      \n\t" // 0
1022                 "psubw %%mm2, %%mm6                     \n\t"
1023                 "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1024                 "movq %%mm7, %%mm6                      \n\t" // 0
1025                 "psubw %%mm3, %%mm6                     \n\t"
1026                 "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1027 #else
1028                 "movq %%mm7, %%mm6                      \n\t" // 0
1029                 "pcmpgtw %%mm0, %%mm6                   \n\t"
1030                 "pxor %%mm6, %%mm0                      \n\t"
1031                 "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1032                 "movq %%mm7, %%mm6                      \n\t" // 0
1033                 "pcmpgtw %%mm1, %%mm6                   \n\t"
1034                 "pxor %%mm6, %%mm1                      \n\t"
1035                 "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1036                 "movq %%mm7, %%mm6                      \n\t" // 0
1037                 "pcmpgtw %%mm2, %%mm6                   \n\t"
1038                 "pxor %%mm6, %%mm2                      \n\t"
1039                 "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1040                 "movq %%mm7, %%mm6                      \n\t" // 0
1041                 "pcmpgtw %%mm3, %%mm6                   \n\t"
1042                 "pxor %%mm6, %%mm3                      \n\t"
1043                 "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1044 #endif
1045
1046 #ifdef HAVE_MMX2
1047                 "pminsw %%mm2, %%mm0                    \n\t"
1048                 "pminsw %%mm3, %%mm1                    \n\t"
1049 #else
1050                 "movq %%mm0, %%mm6                      \n\t"
1051                 "psubusw %%mm2, %%mm6                   \n\t"
1052                 "psubw %%mm6, %%mm0                     \n\t"
1053                 "movq %%mm1, %%mm6                      \n\t"
1054                 "psubusw %%mm3, %%mm6                   \n\t"
1055                 "psubw %%mm6, %%mm1                     \n\t"
1056 #endif
1057
1058                 "movd %2, %%mm2                         \n\t" // QP
1059                 "punpcklbw %%mm7, %%mm2                 \n\t"
1060
1061                 "movq %%mm7, %%mm6                      \n\t" // 0
1062                 "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
1063                 "pxor %%mm6, %%mm4                      \n\t"
1064                 "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
1065                 "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
1066                 "pxor %%mm7, %%mm5                      \n\t"
1067                 "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
1068 // 100 opcodes
1069                 "psllw $3, %%mm2                        \n\t" // 8QP
1070                 "movq %%mm2, %%mm3                      \n\t" // 8QP
1071                 "pcmpgtw %%mm4, %%mm2                   \n\t"
1072                 "pcmpgtw %%mm5, %%mm3                   \n\t"
1073                 "pand %%mm2, %%mm4                      \n\t"
1074                 "pand %%mm3, %%mm5                      \n\t"
1075
1076
1077                 "psubusw %%mm0, %%mm4                   \n\t" // hd
1078                 "psubusw %%mm1, %%mm5                   \n\t" // ld
1079
1080
1081                 "movq "MANGLE(w05)", %%mm2              \n\t" // 5
1082                 "pmullw %%mm2, %%mm4                    \n\t"
1083                 "pmullw %%mm2, %%mm5                    \n\t"
1084                 "movq "MANGLE(w20)", %%mm2              \n\t" // 32
1085                 "paddw %%mm2, %%mm4                     \n\t"
1086                 "paddw %%mm2, %%mm5                     \n\t"
1087                 "psrlw $6, %%mm4                        \n\t"
1088                 "psrlw $6, %%mm5                        \n\t"
1089
1090                 "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
1091                 "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
1092
1093                 "pxor %%mm2, %%mm2                      \n\t"
1094                 "pxor %%mm3, %%mm3                      \n\t"
1095
1096                 "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
1097                 "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
1098                 "pxor %%mm2, %%mm0                      \n\t"
1099                 "pxor %%mm3, %%mm1                      \n\t"
1100                 "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
1101                 "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
1102                 "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
1103                 "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
1104
1105                 "pxor %%mm6, %%mm2                      \n\t"
1106                 "pxor %%mm7, %%mm3                      \n\t"
1107                 "pand %%mm2, %%mm4                      \n\t"
1108                 "pand %%mm3, %%mm5                      \n\t"
1109
1110 #ifdef HAVE_MMX2
1111                 "pminsw %%mm0, %%mm4                    \n\t"
1112                 "pminsw %%mm1, %%mm5                    \n\t"
1113 #else
1114                 "movq %%mm4, %%mm2                      \n\t"
1115                 "psubusw %%mm0, %%mm2                   \n\t"
1116                 "psubw %%mm2, %%mm4                     \n\t"
1117                 "movq %%mm5, %%mm2                      \n\t"
1118                 "psubusw %%mm1, %%mm2                   \n\t"
1119                 "psubw %%mm2, %%mm5                     \n\t"
1120 #endif
1121                 "pxor %%mm6, %%mm4                      \n\t"
1122                 "pxor %%mm7, %%mm5                      \n\t"
1123                 "psubw %%mm6, %%mm4                     \n\t"
1124                 "psubw %%mm7, %%mm5                     \n\t"
1125                 "packsswb %%mm5, %%mm4                  \n\t"
1126                 "movq (%0), %%mm0                       \n\t"
1127                 "paddb   %%mm4, %%mm0                   \n\t"
1128                 "movq %%mm0, (%0)                       \n\t"
1129                 "movq (%0, %1), %%mm0                   \n\t"
1130                 "psubb %%mm4, %%mm0                     \n\t"
1131                 "movq %%mm0, (%0, %1)                   \n\t"
1132
1133                 : "+r" (src)
1134                 : "r" ((long)stride), "m" (c->pQPb)
1135                 : "%"REG_a, "%"REG_c
1136         );
1137 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1138         const int l1= stride;
1139         const int l2= stride + l1;
1140         const int l3= stride + l2;
1141         const int l4= stride + l3;
1142         const int l5= stride + l4;
1143         const int l6= stride + l5;
1144         const int l7= stride + l6;
1145         const int l8= stride + l7;
1146 //        const int l9= stride + l8;
1147         int x;
1148         src+= stride*3;
1149         for(x=0; x<BLOCK_SIZE; x++)
1150         {
1151                 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1152                 if(FFABS(middleEnergy) < 8*c->QP)
1153                 {
1154                         const int q=(src[l4] - src[l5])/2;
1155                         const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1156                         const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1157
1158                         int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
1159                         d= FFMAX(d, 0);
1160
1161                         d= (5*d + 32) >> 6;
1162                         d*= FFSIGN(-middleEnergy);
1163
1164                         if(q>0)
1165                         {
1166                                 d= d<0 ? 0 : d;
1167                                 d= d>q ? q : d;
1168                         }
1169                         else
1170                         {
1171                                 d= d>0 ? 0 : d;
1172                                 d= d<q ? q : d;
1173                         }
1174
1175                         src[l4]-= d;
1176                         src[l5]+= d;
1177                 }
1178                 src++;
1179         }
1180 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1181 }
1182 #endif //HAVE_ALTIVEC
1183
1184 #ifndef HAVE_ALTIVEC
1185 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1186 {
1187 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1188         asm volatile(
1189                 "pxor %%mm6, %%mm6                      \n\t"
1190                 "pcmpeqb %%mm7, %%mm7                   \n\t"
1191                 "movq %2, %%mm0                         \n\t"
1192                 "punpcklbw %%mm6, %%mm0                 \n\t"
1193                 "psrlw $1, %%mm0                        \n\t"
1194                 "psubw %%mm7, %%mm0                     \n\t"
1195                 "packuswb %%mm0, %%mm0                  \n\t"
1196                 "movq %%mm0, %3                         \n\t"
1197
1198                 "lea (%0, %1), %%"REG_a"                \n\t"
1199                 "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1200
1201 //        0        1        2        3        4        5        6        7        8        9
1202 //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1
1203
1204 #undef FIND_MIN_MAX
1205 #ifdef HAVE_MMX2
1206 #define REAL_FIND_MIN_MAX(addr)\
1207                 "movq " #addr ", %%mm0                  \n\t"\
1208                 "pminub %%mm0, %%mm7                    \n\t"\
1209                 "pmaxub %%mm0, %%mm6                    \n\t"
1210 #else
1211 #define REAL_FIND_MIN_MAX(addr)\
1212                 "movq " #addr ", %%mm0                  \n\t"\
1213                 "movq %%mm7, %%mm1                      \n\t"\
1214                 "psubusb %%mm0, %%mm6                   \n\t"\
1215                 "paddb %%mm0, %%mm6                     \n\t"\
1216                 "psubusb %%mm0, %%mm1                   \n\t"\
1217                 "psubb %%mm1, %%mm7                     \n\t"
1218 #endif
1219 #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
1220
1221 FIND_MIN_MAX((%%REGa))
1222 FIND_MIN_MAX((%%REGa, %1))
1223 FIND_MIN_MAX((%%REGa, %1, 2))
1224 FIND_MIN_MAX((%0, %1, 4))
1225 FIND_MIN_MAX((%%REGd))
1226 FIND_MIN_MAX((%%REGd, %1))
1227 FIND_MIN_MAX((%%REGd, %1, 2))
1228 FIND_MIN_MAX((%0, %1, 8))
1229
1230                 "movq %%mm7, %%mm4                      \n\t"
1231                 "psrlq $8, %%mm7                        \n\t"
1232 #ifdef HAVE_MMX2
1233                 "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1234                 "pshufw $0xF9, %%mm7, %%mm4             \n\t"
1235                 "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1236                 "pshufw $0xFE, %%mm7, %%mm4             \n\t"
1237                 "pminub %%mm4, %%mm7                    \n\t"
1238 #else
1239                 "movq %%mm7, %%mm1                      \n\t"
1240                 "psubusb %%mm4, %%mm1                   \n\t"
1241                 "psubb %%mm1, %%mm7                     \n\t"
1242                 "movq %%mm7, %%mm4                      \n\t"
1243                 "psrlq $16, %%mm7                       \n\t"
1244                 "movq %%mm7, %%mm1                      \n\t"
1245                 "psubusb %%mm4, %%mm1                   \n\t"
1246                 "psubb %%mm1, %%mm7                     \n\t"
1247                 "movq %%mm7, %%mm4                      \n\t"
1248                 "psrlq $32, %%mm7                       \n\t"
1249                 "movq %%mm7, %%mm1                      \n\t"
1250                 "psubusb %%mm4, %%mm1                   \n\t"
1251                 "psubb %%mm1, %%mm7                     \n\t"
1252 #endif
1253
1254
1255                 "movq %%mm6, %%mm4                      \n\t"
1256                 "psrlq $8, %%mm6                        \n\t"
1257 #ifdef HAVE_MMX2
1258                 "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
1259                 "pshufw $0xF9, %%mm6, %%mm4             \n\t"
1260                 "pmaxub %%mm4, %%mm6                    \n\t"
1261                 "pshufw $0xFE, %%mm6, %%mm4             \n\t"
1262                 "pmaxub %%mm4, %%mm6                    \n\t"
1263 #else
1264                 "psubusb %%mm4, %%mm6                   \n\t"
1265                 "paddb %%mm4, %%mm6                     \n\t"
1266                 "movq %%mm6, %%mm4                      \n\t"
1267                 "psrlq $16, %%mm6                       \n\t"
1268                 "psubusb %%mm4, %%mm6                   \n\t"
1269                 "paddb %%mm4, %%mm6                     \n\t"
1270                 "movq %%mm6, %%mm4                      \n\t"
1271                 "psrlq $32, %%mm6                       \n\t"
1272                 "psubusb %%mm4, %%mm6                   \n\t"
1273                 "paddb %%mm4, %%mm6                     \n\t"
1274 #endif
1275                 "movq %%mm6, %%mm0                      \n\t" // max
1276                 "psubb %%mm7, %%mm6                     \n\t" // max - min
1277                 "movd %%mm6, %%ecx                      \n\t"
1278                 "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
1279                 " jb 1f                                 \n\t"
1280                 "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
1281                 "and "ALIGN_MASK", %%"REG_c"            \n\t"
1282                 PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
1283                 "punpcklbw %%mm7, %%mm7                 \n\t"
1284                 "punpcklbw %%mm7, %%mm7                 \n\t"
1285                 "punpcklbw %%mm7, %%mm7                 \n\t"
1286                 "movq %%mm7, (%%"REG_c")                \n\t"
1287
1288                 "movq (%0), %%mm0                       \n\t" // L10
1289                 "movq %%mm0, %%mm1                      \n\t" // L10
1290                 "movq %%mm0, %%mm2                      \n\t" // L10
1291                 "psllq $8, %%mm1                        \n\t"
1292                 "psrlq $8, %%mm2                        \n\t"
1293                 "movd -4(%0), %%mm3                     \n\t"
1294                 "movd 8(%0), %%mm4                      \n\t"
1295                 "psrlq $24, %%mm3                       \n\t"
1296                 "psllq $56, %%mm4                       \n\t"
1297                 "por %%mm3, %%mm1                       \n\t" // L00
1298                 "por %%mm4, %%mm2                       \n\t" // L20
1299                 "movq %%mm1, %%mm3                      \n\t" // L00
1300                 PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
1301                 PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
1302                 "psubusb %%mm7, %%mm0                   \n\t"
1303                 "psubusb %%mm7, %%mm2                   \n\t"
1304                 "psubusb %%mm7, %%mm3                   \n\t"
1305                 "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
1306                 "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
1307                 "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
1308                 "paddb %%mm2, %%mm0                     \n\t"
1309                 "paddb %%mm3, %%mm0                     \n\t"
1310
1311                 "movq (%%"REG_a"), %%mm2                \n\t" // L11
1312                 "movq %%mm2, %%mm3                      \n\t" // L11
1313                 "movq %%mm2, %%mm4                      \n\t" // L11
1314                 "psllq $8, %%mm3                        \n\t"
1315                 "psrlq $8, %%mm4                        \n\t"
1316                 "movd -4(%%"REG_a"), %%mm5              \n\t"
1317                 "movd 8(%%"REG_a"), %%mm6               \n\t"
1318                 "psrlq $24, %%mm5                       \n\t"
1319                 "psllq $56, %%mm6                       \n\t"
1320                 "por %%mm5, %%mm3                       \n\t" // L01
1321                 "por %%mm6, %%mm4                       \n\t" // L21
1322                 "movq %%mm3, %%mm5                      \n\t" // L01
1323                 PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
1324                 PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
1325                 "psubusb %%mm7, %%mm2                   \n\t"
1326                 "psubusb %%mm7, %%mm4                   \n\t"
1327                 "psubusb %%mm7, %%mm5                   \n\t"
1328                 "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
1329                 "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
1330                 "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
1331                 "paddb %%mm4, %%mm2                     \n\t"
1332                 "paddb %%mm5, %%mm2                     \n\t"
1333 // 0, 2, 3, 1
1334 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1335                 "movq " #src ", " #sx "                 \n\t" /* src[0] */\
1336                 "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
1337                 "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
1338                 "psllq $8, " #lx "                      \n\t"\
1339                 "psrlq $8, " #t0 "                      \n\t"\
1340                 "movd -4" #src ", " #t1 "               \n\t"\
1341                 "psrlq $24, " #t1 "                     \n\t"\
1342                 "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
1343                 "movd 8" #src ", " #t1 "                \n\t"\
1344                 "psllq $56, " #t1 "                     \n\t"\
1345                 "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
1346                 "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
1347                 PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
1348                 PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
1349                 PAVGB(lx, pplx)                                     \
1350                 "movq " #lx ", 8(%%"REG_c")             \n\t"\
1351                 "movq (%%"REG_c"), " #lx "              \n\t"\
1352                 "psubusb " #lx ", " #t1 "               \n\t"\
1353                 "psubusb " #lx ", " #t0 "               \n\t"\
1354                 "psubusb " #lx ", " #sx "               \n\t"\
1355                 "movq "MANGLE(b00)", " #lx "            \n\t"\
1356                 "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
1357                 "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
1358                 "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
1359                 "paddb " #t1 ", " #t0 "                 \n\t"\
1360                 "paddb " #t0 ", " #sx "                 \n\t"\
1361 \
1362                 PAVGB(plx, pplx)                              /* filtered */\
1363                 "movq " #dst ", " #t0 "                 \n\t" /* dst */\
1364                 "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
1365                 "psubusb %3, " #t0 "                    \n\t"\
1366                 "paddusb %3, " #t1 "                    \n\t"\
1367                 PMAXUB(t0, pplx)\
1368                 PMINUB(t1, pplx, t0)\
1369                 "paddb " #sx ", " #ppsx "               \n\t"\
1370                 "paddb " #psx ", " #ppsx "              \n\t"\
1371                 "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
1372                 "pand "MANGLE(b08)", " #ppsx "          \n\t"\
1373                 "pcmpeqb " #lx ", " #ppsx "             \n\t"\
1374                 "pand " #ppsx ", " #pplx "              \n\t"\
1375                 "pandn " #dst ", " #ppsx "              \n\t"\
1376                 "por " #pplx ", " #ppsx "               \n\t"\
1377                 "movq " #ppsx ", " #dst "               \n\t"\
1378                 "movq 8(%%"REG_c"), " #lx "             \n\t"
1379
1380 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1381    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
1382 /*
1383 0000000
1384 1111111
1385
1386 1111110
1387 1111101
1388 1111100
1389 1111011
1390 1111010
1391 1111001
1392
1393 1111000
1394 1110111
1395
1396 */
1397 //DERING_CORE(dst          ,src            ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)
1398 DERING_CORE((%%REGa)       ,(%%REGa, %1)   ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1399 DERING_CORE((%%REGa, %1)   ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1400 DERING_CORE((%%REGa, %1, 2),(%0, %1, 4)    ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1401 DERING_CORE((%0, %1, 4)    ,(%%REGd)       ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1402 DERING_CORE((%%REGd)       ,(%%REGd, %1)   ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1403 DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1404 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1405 DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1406
1407                 "1:                        \n\t"
1408                 : : "r" (src), "r" ((long)stride), "m" (c->pQPb), "m"(c->pQPb2)
1409                 : "%"REG_a, "%"REG_d, "%"REG_c
1410         );
1411 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1412         int y;
1413         int min=255;
1414         int max=0;
1415         int avg;
1416         uint8_t *p;
1417         int s[10];
1418         const int QP2= c->QP/2 + 1;
1419
1420         for(y=1; y<9; y++)
1421         {
1422                 int x;
1423                 p= src + stride*y;
1424                 for(x=1; x<9; x++)
1425                 {
1426                         p++;
1427                         if(*p > max) max= *p;
1428                         if(*p < min) min= *p;
1429                 }
1430         }
1431         avg= (min + max + 1)>>1;
1432
1433         if(max - min <deringThreshold) return;
1434
1435         for(y=0; y<10; y++)
1436         {
1437                 int t = 0;
1438
1439                 if(src[stride*y + 0] > avg) t+= 1;
1440                 if(src[stride*y + 1] > avg) t+= 2;
1441                 if(src[stride*y + 2] > avg) t+= 4;
1442                 if(src[stride*y + 3] > avg) t+= 8;
1443                 if(src[stride*y + 4] > avg) t+= 16;
1444                 if(src[stride*y + 5] > avg) t+= 32;
1445                 if(src[stride*y + 6] > avg) t+= 64;
1446                 if(src[stride*y + 7] > avg) t+= 128;
1447                 if(src[stride*y + 8] > avg) t+= 256;
1448                 if(src[stride*y + 9] > avg) t+= 512;
1449
1450                 t |= (~t)<<16;
1451                 t &= (t<<1) & (t>>1);
1452                 s[y] = t;
1453         }
1454
1455         for(y=1; y<9; y++)
1456         {
1457                 int t = s[y-1] & s[y] & s[y+1];
1458                 t|= t>>16;
1459                 s[y-1]= t;
1460         }
1461
1462         for(y=1; y<9; y++)
1463         {
1464                 int x;
1465                 int t = s[y-1];
1466
1467                 p= src + stride*y;
1468                 for(x=1; x<9; x++)
1469                 {
1470                         p++;
1471                         if(t & (1<<x))
1472                         {
1473                                 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1474                                       +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
1475                                       +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1476                                 f= (f + 8)>>4;
1477
1478 #ifdef DEBUG_DERING_THRESHOLD
1479                                 asm volatile("emms\n\t":);
1480                                 {
1481                                 static long long numPixels=0;
1482                                 if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1483 //                                if((max-min)<20 || (max-min)*QP<200)
1484 //                                if((max-min)*QP < 500)
1485 //                                if(max-min<QP/2)
1486                                 if(max-min < 20)
1487                                 {
1488                                         static int numSkiped=0;
1489                                         static int errorSum=0;
1490                                         static int worstQP=0;
1491                                         static int worstRange=0;
1492                                         static int worstDiff=0;
1493                                         int diff= (f - *p);
1494                                         int absDiff= FFABS(diff);
1495                                         int error= diff*diff;
1496
1497                                         if(x==1 || x==8 || y==1 || y==8) continue;
1498
1499                                         numSkiped++;
1500                                         if(absDiff > worstDiff)
1501                                         {
1502                                                 worstDiff= absDiff;
1503                                                 worstQP= QP;
1504                                                 worstRange= max-min;
1505                                         }
1506                                         errorSum+= error;
1507
1508                                         if(1024LL*1024LL*1024LL % numSkiped == 0)
1509                                         {
1510                                                 av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
1511                                                         "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1512                                                         (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
1513                                                         worstDiff, (float)numSkiped/numPixels);
1514                                         }
1515                                 }
1516                                 }
1517 #endif
1518                                 if     (*p + QP2 < f) *p= *p + QP2;
1519                                 else if(*p - QP2 > f) *p= *p - QP2;
1520                                 else *p=f;
1521                         }
1522                 }
1523         }
1524 #ifdef DEBUG_DERING_THRESHOLD
1525         if(max-min < 20)
1526         {
1527                 for(y=1; y<9; y++)
1528                 {
1529                         int x;
1530                         int t = 0;
1531                         p= src + stride*y;
1532                         for(x=1; x<9; x++)
1533                         {
1534                                 p++;
1535                                 *p = FFMIN(*p + 20, 255);
1536                         }
1537                 }
1538 //                src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1539         }
1540 #endif
1541 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1542 }
1543 #endif //HAVE_ALTIVEC
1544
1545 /**
1546  * Deinterlaces the given block by linearly interpolating every second line.
1547  * will be called for every 8x8 block and can read & write from line 4-15
1548  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1549  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1550  */
1551 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
1552 {
1553 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1554         src+= 4*stride;
1555         asm volatile(
1556                 "lea (%0, %1), %%"REG_a"                \n\t"
1557                 "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
1558 //      0       1       2       3       4       5       6       7       8       9
1559 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
1560
1561                 "movq (%0), %%mm0                       \n\t"
1562                 "movq (%%"REG_a", %1), %%mm1            \n\t"
1563                 PAVGB(%%mm1, %%mm0)
1564                 "movq %%mm0, (%%"REG_a")                \n\t"
1565                 "movq (%0, %1, 4), %%mm0                \n\t"
1566                 PAVGB(%%mm0, %%mm1)
1567                 "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
1568                 "movq (%%"REG_c", %1), %%mm1            \n\t"
1569                 PAVGB(%%mm1, %%mm0)
1570                 "movq %%mm0, (%%"REG_c")                \n\t"
1571                 "movq (%0, %1, 8), %%mm0                \n\t"
1572                 PAVGB(%%mm0, %%mm1)
1573                 "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
1574
1575                 : : "r" (src), "r" ((long)stride)
1576                 : "%"REG_a, "%"REG_c
1577         );
1578 #else
1579         int a, b, x;
1580         src+= 4*stride;
1581
1582         for(x=0; x<2; x++){
1583                 a= *(uint32_t*)&src[stride*0];
1584                 b= *(uint32_t*)&src[stride*2];
1585                 *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1586                 a= *(uint32_t*)&src[stride*4];
1587                 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1588                 b= *(uint32_t*)&src[stride*6];
1589                 *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1590                 a= *(uint32_t*)&src[stride*8];
1591                 *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1592                 src += 4;
1593         }
1594 #endif
1595 }
1596
1597 /**
1598  * Deinterlaces the given block by cubic interpolating every second line.
1599  * will be called for every 8x8 block and can read & write from line 4-15
1600  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1601  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1602  * this filter will read lines 3-15 and write 7-13
1603  */
1604 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
1605 {
1606 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1607         src+= stride*3;
1608         asm volatile(
1609                 "lea (%0, %1), %%"REG_a"                \n\t"
1610                 "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1611                 "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
1612                 "add %1, %%"REG_c"                      \n\t"
1613                 "pxor %%mm7, %%mm7                      \n\t"
1614 //      0       1       2       3       4       5       6       7       8       9       10
1615 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1616
1617 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
1618                 "movq " #a ", %%mm0                     \n\t"\
1619                 "movq " #b ", %%mm1                     \n\t"\
1620                 "movq " #d ", %%mm2                     \n\t"\
1621                 "movq " #e ", %%mm3                     \n\t"\
1622                 PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
1623                 PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
1624                 "movq %%mm0, %%mm2                      \n\t"\
1625                 "punpcklbw %%mm7, %%mm0                 \n\t"\
1626                 "punpckhbw %%mm7, %%mm2                 \n\t"\
1627                 "movq %%mm1, %%mm3                      \n\t"\
1628                 "punpcklbw %%mm7, %%mm1                 \n\t"\
1629                 "punpckhbw %%mm7, %%mm3                 \n\t"\
1630                 "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
1631                 "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
1632                 "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
1633                 "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
1634                 "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
1635                 "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
1636                 "packuswb %%mm3, %%mm1                  \n\t"\
1637                 "movq %%mm1, " #c "                     \n\t"
1638 #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)
1639
1640 DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
1641 DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))
1642 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
1643 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
1644
1645                 : : "r" (src), "r" ((long)stride)
1646                 : "%"REG_a, "%"REG_d, "%"REG_c
1647         );
1648 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1649         int x;
1650         src+= stride*3;
1651         for(x=0; x<8; x++)
1652         {
1653                 src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1654                 src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1655                 src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1656                 src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1657                 src++;
1658         }
1659 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1660 }
1661
1662 /**
1663  * Deinterlaces the given block by filtering every second line with a (-1 4 2 4 -1) filter.
1664  * will be called for every 8x8 block and can read & write from line 4-15
1665  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1666  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1667  * this filter will read lines 4-13 and write 5-11
1668  */
1669 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
1670 {
1671 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1672         src+= stride*4;
1673         asm volatile(
1674                 "lea (%0, %1), %%"REG_a"                \n\t"
1675                 "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1676                 "pxor %%mm7, %%mm7                      \n\t"
1677                 "movq (%2), %%mm0                       \n\t"
1678 //      0       1       2       3       4       5       6       7       8       9       10
1679 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1680
1681 #define REAL_DEINT_FF(a,b,c,d)\
1682                 "movq " #a ", %%mm1                     \n\t"\
1683                 "movq " #b ", %%mm2                     \n\t"\
1684                 "movq " #c ", %%mm3                     \n\t"\
1685                 "movq " #d ", %%mm4                     \n\t"\
1686                 PAVGB(%%mm3, %%mm1)                          \
1687                 PAVGB(%%mm4, %%mm0)                          \
1688                 "movq %%mm0, %%mm3                      \n\t"\
1689                 "punpcklbw %%mm7, %%mm0                 \n\t"\
1690                 "punpckhbw %%mm7, %%mm3                 \n\t"\
1691                 "movq %%mm1, %%mm4                      \n\t"\
1692                 "punpcklbw %%mm7, %%mm1                 \n\t"\
1693                 "punpckhbw %%mm7, %%mm4                 \n\t"\
1694                 "psllw $2, %%mm1                        \n\t"\
1695                 "psllw $2, %%mm4                        \n\t"\
1696                 "psubw %%mm0, %%mm1                     \n\t"\
1697                 "psubw %%mm3, %%mm4                     \n\t"\
1698                 "movq %%mm2, %%mm5                      \n\t"\
1699                 "movq %%mm2, %%mm0                      \n\t"\
1700                 "punpcklbw %%mm7, %%mm2                 \n\t"\
1701                 "punpckhbw %%mm7, %%mm5                 \n\t"\
1702                 "paddw %%mm2, %%mm1                     \n\t"\
1703                 "paddw %%mm5, %%mm4                     \n\t"\
1704                 "psraw $2, %%mm1                        \n\t"\
1705                 "psraw $2, %%mm4                        \n\t"\
1706                 "packuswb %%mm4, %%mm1                  \n\t"\
1707                 "movq %%mm1, " #b "                     \n\t"\
1708
1709 #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)
1710
1711 DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))
1712 DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
1713 DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
1714 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1715
1716                 "movq %%mm0, (%2)                       \n\t"
1717                 : : "r" (src), "r" ((long)stride), "r"(tmp)
1718                 : "%"REG_a, "%"REG_d
1719         );
1720 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1721         int x;
1722         src+= stride*4;
1723         for(x=0; x<8; x++)
1724         {
1725                 int t1= tmp[x];
1726                 int t2= src[stride*1];
1727
1728                 src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1729                 t1= src[stride*4];
1730                 src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1731                 t2= src[stride*6];
1732                 src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1733                 t1= src[stride*8];
1734                 src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1735                 tmp[x]= t1;
1736
1737                 src++;
1738         }
1739 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1740 }
1741
1742 /**
1743  * Deinterlaces the given block by filtering every line with a (-1 2 6 2 -1) filter.
1744  * will be called for every 8x8 block and can read & write from line 4-15
1745  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1746  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1747  * this filter will read lines 4-13 and write 4-11
1748  */
1749 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
1750 {
1751 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1752         src+= stride*4;
1753         asm volatile(
1754                 "lea (%0, %1), %%"REG_a"                \n\t"
1755                 "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1756                 "pxor %%mm7, %%mm7                      \n\t"
1757                 "movq (%2), %%mm0                       \n\t"
1758                 "movq (%3), %%mm1                       \n\t"
1759 //      0       1       2       3       4       5       6       7       8       9       10
1760 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1761
1762 #define REAL_DEINT_L5(t1,t2,a,b,c)\
1763                 "movq " #a ", %%mm2                     \n\t"\
1764                 "movq " #b ", %%mm3                     \n\t"\
1765                 "movq " #c ", %%mm4                     \n\t"\
1766                 PAVGB(t2, %%mm3)                             \
1767                 PAVGB(t1, %%mm4)                             \
1768                 "movq %%mm2, %%mm5                      \n\t"\
1769                 "movq %%mm2, " #t1 "                    \n\t"\
1770                 "punpcklbw %%mm7, %%mm2                 \n\t"\
1771                 "punpckhbw %%mm7, %%mm5                 \n\t"\
1772                 "movq %%mm2, %%mm6                      \n\t"\
1773                 "paddw %%mm2, %%mm2                     \n\t"\
1774                 "paddw %%mm6, %%mm2                     \n\t"\
1775                 "movq %%mm5, %%mm6                      \n\t"\
1776                 "paddw %%mm5, %%mm5                     \n\t"\
1777                 "paddw %%mm6, %%mm5                     \n\t"\
1778                 "movq %%mm3, %%mm6                      \n\t"\
1779                 "punpcklbw %%mm7, %%mm3                 \n\t"\
1780                 "punpckhbw %%mm7, %%mm6                 \n\t"\
1781                 "paddw %%mm3, %%mm3                     \n\t"\
1782                 "paddw %%mm6, %%mm6                     \n\t"\
1783                 "paddw %%mm3, %%mm2                     \n\t"\
1784                 "paddw %%mm6, %%mm5                     \n\t"\
1785                 "movq %%mm4, %%mm6                      \n\t"\
1786                 "punpcklbw %%mm7, %%mm4                 \n\t"\
1787                 "punpckhbw %%mm7, %%mm6                 \n\t"\
1788                 "psubw %%mm4, %%mm2                     \n\t"\
1789                 "psubw %%mm6, %%mm5                     \n\t"\
1790                 "psraw $2, %%mm2                        \n\t"\
1791                 "psraw $2, %%mm5                        \n\t"\
1792                 "packuswb %%mm5, %%mm2                  \n\t"\
1793                 "movq %%mm2, " #a "                     \n\t"\
1794
1795 #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)
1796
1797 DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )
1798 DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))
1799 DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )
1800 DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )
1801 DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )
1802 DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
1803 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
1804 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
1805
1806                 "movq %%mm0, (%2)                       \n\t"
1807                 "movq %%mm1, (%3)                       \n\t"
1808                 : : "r" (src), "r" ((long)stride), "r"(tmp), "r"(tmp2)
1809                 : "%"REG_a, "%"REG_d
1810         );
1811 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1812         int x;
1813         src+= stride*4;
1814         for(x=0; x<8; x++)
1815         {
1816                 int t1= tmp[x];
1817                 int t2= tmp2[x];
1818                 int t3= src[0];
1819
1820                 src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1821                 t1= src[stride*1];
1822                 src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1823                 t2= src[stride*2];
1824                 src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1825                 t3= src[stride*3];
1826                 src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1827                 t1= src[stride*4];
1828                 src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1829                 t2= src[stride*5];
1830                 src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1831                 t3= src[stride*6];
1832                 src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1833                 t1= src[stride*7];
1834                 src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1835
1836                 tmp[x]= t3;
1837                 tmp2[x]= t1;
1838
1839                 src++;
1840         }
1841 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1842 }
1843
1844 /**
1845  * Deinterlaces the given block by filtering all lines with a (1 2 1) filter.
1846  * will be called for every 8x8 block and can read & write from line 4-15
1847  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1848  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1849  * this filter will read lines 4-13 and write 4-11
1850  */
1851 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
1852 {
1853 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1854         src+= 4*stride;
1855         asm volatile(
1856                 "lea (%0, %1), %%"REG_a"                \n\t"
1857                 "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1858 //      0       1       2       3       4       5       6       7       8       9
1859 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1860
1861                 "movq (%2), %%mm0                       \n\t" // L0
1862                 "movq (%%"REG_a"), %%mm1                \n\t" // L2
1863                 PAVGB(%%mm1, %%mm0)                           // L0+L2
1864                 "movq (%0), %%mm2                       \n\t" // L1
1865                 PAVGB(%%mm2, %%mm0)
1866                 "movq %%mm0, (%0)                       \n\t"
1867                 "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
1868                 PAVGB(%%mm0, %%mm2)                           // L1+L3
1869                 PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
1870                 "movq %%mm2, (%%"REG_a")                \n\t"
1871                 "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
1872                 PAVGB(%%mm2, %%mm1)                           // L2+L4
1873                 PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
1874                 "movq %%mm1, (%%"REG_a", %1)            \n\t"
1875                 "movq (%0, %1, 4), %%mm1                \n\t" // L5
1876                 PAVGB(%%mm1, %%mm0)                           // L3+L5
1877                 PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
1878                 "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
1879                 "movq (%%"REG_d"), %%mm0                \n\t" // L6
1880                 PAVGB(%%mm0, %%mm2)                           // L4+L6
1881                 PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
1882                 "movq %%mm2, (%0, %1, 4)                \n\t"
1883                 "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
1884                 PAVGB(%%mm2, %%mm1)                           // L5+L7
1885                 PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
1886                 "movq %%mm1, (%%"REG_d")                \n\t"
1887                 "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
1888                 PAVGB(%%mm1, %%mm0)                           // L6+L8
1889                 PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
1890                 "movq %%mm0, (%%"REG_d", %1)            \n\t"
1891                 "movq (%0, %1, 8), %%mm0                \n\t" // L9
1892                 PAVGB(%%mm0, %%mm2)                           // L7+L9
1893                 PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
1894                 "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
1895                 "movq %%mm1, (%2)                       \n\t"
1896
1897                 : : "r" (src), "r" ((long)stride), "r" (tmp)
1898                 : "%"REG_a, "%"REG_d
1899         );
1900 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1901         int a, b, c, x;
1902         src+= 4*stride;
1903
1904         for(x=0; x<2; x++){
1905                 a= *(uint32_t*)&tmp[stride*0];
1906                 b= *(uint32_t*)&src[stride*0];
1907                 c= *(uint32_t*)&src[stride*1];
1908                 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1909                 *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1910
1911                 a= *(uint32_t*)&src[stride*2];
1912                 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1913                 *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1914
1915                 b= *(uint32_t*)&src[stride*3];
1916                 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1917                 *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1918
1919                 c= *(uint32_t*)&src[stride*4];
1920                 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1921                 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1922
1923                 a= *(uint32_t*)&src[stride*5];
1924                 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1925                 *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1926
1927                 b= *(uint32_t*)&src[stride*6];
1928                 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1929                 *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1930
1931                 c= *(uint32_t*)&src[stride*7];
1932                 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1933                 *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1934
1935                 a= *(uint32_t*)&src[stride*8];
1936                 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1937                 *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1938
1939                 *(uint32_t*)&tmp[stride*0]= c;
1940                 src += 4;
1941                 tmp += 4;
1942         }
1943 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1944 }
1945
1946 /**
1947  * Deinterlaces the given block by applying a median filter to every second line.
1948  * will be called for every 8x8 block and can read & write from line 4-15,
1949  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1950  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1951  */
1952 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
1953 {
1954 #ifdef HAVE_MMX
1955         src+= 4*stride;
1956 #ifdef HAVE_MMX2
1957         asm volatile(
1958                 "lea (%0, %1), %%"REG_a"                \n\t"
1959                 "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1960 //      0       1       2       3       4       5       6       7       8       9
1961 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1962
1963                 "movq (%0), %%mm0                       \n\t" //
1964                 "movq (%%"REG_a", %1), %%mm2            \n\t" //
1965                 "movq (%%"REG_a"), %%mm1                \n\t" //
1966                 "movq %%mm0, %%mm3                      \n\t"
1967                 "pmaxub %%mm1, %%mm0                    \n\t" //
1968                 "pminub %%mm3, %%mm1                    \n\t" //
1969                 "pmaxub %%mm2, %%mm1                    \n\t" //
1970                 "pminub %%mm1, %%mm0                    \n\t"
1971                 "movq %%mm0, (%%"REG_a")                \n\t"
1972
1973                 "movq (%0, %1, 4), %%mm0                \n\t" //
1974                 "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
1975                 "movq %%mm2, %%mm3                      \n\t"
1976                 "pmaxub %%mm1, %%mm2                    \n\t" //
1977                 "pminub %%mm3, %%mm1                    \n\t" //
1978                 "pmaxub %%mm0, %%mm1                    \n\t" //
1979                 "pminub %%mm1, %%mm2                    \n\t"
1980                 "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
1981
1982                 "movq (%%"REG_d"), %%mm2                \n\t" //
1983                 "movq (%%"REG_d", %1), %%mm1            \n\t" //
1984                 "movq %%mm2, %%mm3                      \n\t"
1985                 "pmaxub %%mm0, %%mm2                    \n\t" //
1986                 "pminub %%mm3, %%mm0                    \n\t" //
1987                 "pmaxub %%mm1, %%mm0                    \n\t" //
1988                 "pminub %%mm0, %%mm2                    \n\t"
1989                 "movq %%mm2, (%%"REG_d")                \n\t"
1990
1991                 "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
1992                 "movq (%0, %1, 8), %%mm0                \n\t" //
1993                 "movq %%mm2, %%mm3                      \n\t"
1994                 "pmaxub %%mm0, %%mm2                    \n\t" //
1995                 "pminub %%mm3, %%mm0                    \n\t" //
1996                 "pmaxub %%mm1, %%mm0                    \n\t" //
1997                 "pminub %%mm0, %%mm2                    \n\t"
1998                 "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
1999
2000
2001                 : : "r" (src), "r" ((long)stride)
2002                 : "%"REG_a, "%"REG_d
2003         );
2004
2005 #else // MMX without MMX2
2006         asm volatile(
2007                 "lea (%0, %1), %%"REG_a"                \n\t"
2008                 "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
2009 //      0       1       2       3       4       5       6       7       8       9
2010 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
2011                 "pxor %%mm7, %%mm7                      \n\t"
2012
2013 #define REAL_MEDIAN(a,b,c)\
2014                 "movq " #a ", %%mm0                     \n\t"\
2015                 "movq " #b ", %%mm2                     \n\t"\
2016                 "movq " #c ", %%mm1                     \n\t"\
2017                 "movq %%mm0, %%mm3                      \n\t"\
2018                 "movq %%mm1, %%mm4                      \n\t"\
2019                 "movq %%mm2, %%mm5                      \n\t"\
2020                 "psubusb %%mm1, %%mm3                   \n\t"\
2021                 "psubusb %%mm2, %%mm4                   \n\t"\
2022                 "psubusb %%mm0, %%mm5                   \n\t"\
2023                 "pcmpeqb %%mm7, %%mm3                   \n\t"\
2024                 "pcmpeqb %%mm7, %%mm4                   \n\t"\
2025                 "pcmpeqb %%mm7, %%mm5                   \n\t"\
2026                 "movq %%mm3, %%mm6                      \n\t"\
2027                 "pxor %%mm4, %%mm3                      \n\t"\
2028                 "pxor %%mm5, %%mm4                      \n\t"\
2029                 "pxor %%mm6, %%mm5                      \n\t"\
2030                 "por %%mm3, %%mm1                       \n\t"\
2031                 "por %%mm4, %%mm2                       \n\t"\
2032                 "por %%mm5, %%mm0                       \n\t"\
2033                 "pand %%mm2, %%mm0                      \n\t"\
2034                 "pand %%mm1, %%mm0                      \n\t"\
2035                 "movq %%mm0, " #b "                     \n\t"
2036 #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
2037
2038 MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))
2039 MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
2040 MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))
2041 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
2042
2043                 : : "r" (src), "r" ((long)stride)
2044                 : "%"REG_a, "%"REG_d
2045         );
2046 #endif //HAVE_MMX2
2047 #else //HAVE_MMX
2048         int x, y;
2049         src+= 4*stride;
2050         // FIXME - there should be a way to do a few columns in parallel like w/mmx
2051         for(x=0; x<8; x++)
2052         {
2053                 uint8_t *colsrc = src;
2054                 for (y=0; y<4; y++)
2055                 {
2056                         int a, b, c, d, e, f;
2057                         a = colsrc[0       ];
2058                         b = colsrc[stride  ];
2059                         c = colsrc[stride*2];
2060                         d = (a-b)>>31;
2061                         e = (b-c)>>31;
2062                         f = (c-a)>>31;
2063                         colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
2064                         colsrc += stride*2;
2065                 }
2066                 src++;
2067         }
2068 #endif //HAVE_MMX
2069 }
2070
2071 #ifdef HAVE_MMX
2072 /**
2073  * transposes and shift the given 8x8 Block into dst1 and dst2
2074  */
2075 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
2076 {
2077         asm(
2078                 "lea (%0, %1), %%"REG_a"                \n\t"
2079 //      0       1       2       3       4       5       6       7       8       9
2080 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
2081                 "movq (%0), %%mm0                       \n\t" // 12345678
2082                 "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2083                 "movq %%mm0, %%mm2                      \n\t" // 12345678
2084                 "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2085                 "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2086
2087                 "movq (%%"REG_a", %1), %%mm1            \n\t"
2088                 "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2089                 "movq %%mm1, %%mm4                      \n\t"
2090                 "punpcklbw %%mm3, %%mm1                 \n\t"
2091                 "punpckhbw %%mm3, %%mm4                 \n\t"
2092
2093                 "movq %%mm0, %%mm3                      \n\t"
2094                 "punpcklwd %%mm1, %%mm0                 \n\t"
2095                 "punpckhwd %%mm1, %%mm3                 \n\t"
2096                 "movq %%mm2, %%mm1                      \n\t"
2097                 "punpcklwd %%mm4, %%mm2                 \n\t"
2098                 "punpckhwd %%mm4, %%mm1                 \n\t"
2099
2100                 "movd %%mm0, 128(%2)                    \n\t"
2101                 "psrlq $32, %%mm0                       \n\t"
2102                 "movd %%mm0, 144(%2)                    \n\t"
2103                 "movd %%mm3, 160(%2)                    \n\t"
2104                 "psrlq $32, %%mm3                       \n\t"
2105                 "movd %%mm3, 176(%2)                    \n\t"
2106                 "movd %%mm3, 48(%3)                     \n\t"
2107                 "movd %%mm2, 192(%2)                    \n\t"
2108                 "movd %%mm2, 64(%3)                     \n\t"
2109                 "psrlq $32, %%mm2                       \n\t"
2110                 "movd %%mm2, 80(%3)                     \n\t"
2111                 "movd %%mm1, 96(%3)                     \n\t"
2112                 "psrlq $32, %%mm1                       \n\t"
2113                 "movd %%mm1, 112(%3)                    \n\t"
2114
2115                 "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
2116
2117                 "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
2118                 "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2119                 "movq %%mm0, %%mm2                      \n\t" // 12345678
2120                 "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2121                 "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2122
2123                 "movq (%%"REG_a", %1), %%mm1            \n\t"
2124                 "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2125                 "movq %%mm1, %%mm4                      \n\t"
2126                 "punpcklbw %%mm3, %%mm1                 \n\t"
2127                 "punpckhbw %%mm3, %%mm4                 \n\t"
2128
2129                 "movq %%mm0, %%mm3                      \n\t"
2130                 "punpcklwd %%mm1, %%mm0                 \n\t"
2131                 "punpckhwd %%mm1, %%mm3                 \n\t"
2132                 "movq %%mm2, %%mm1                      \n\t"
2133                 "punpcklwd %%mm4, %%mm2                 \n\t"
2134                 "punpckhwd %%mm4, %%mm1                 \n\t"
2135
2136                 "movd %%mm0, 132(%2)                    \n\t"
2137                 "psrlq $32, %%mm0                       \n\t"
2138                 "movd %%mm0, 148(%2)                    \n\t"
2139                 "movd %%mm3, 164(%2)                    \n\t"
2140                 "psrlq $32, %%mm3                       \n\t"
2141                 "movd %%mm3, 180(%2)                    \n\t"
2142                 "movd %%mm3, 52(%3)                     \n\t"
2143                 "movd %%mm2, 196(%2)                    \n\t"
2144                 "movd %%mm2, 68(%3)                     \n\t"
2145                 "psrlq $32, %%mm2                       \n\t"
2146                 "movd %%mm2, 84(%3)                     \n\t"
2147                 "movd %%mm1, 100(%3)                    \n\t"
2148                 "psrlq $32, %%mm1                       \n\t"
2149                 "movd %%mm1, 116(%3)                    \n\t"
2150
2151
2152         :: "r" (src), "r" ((long)srcStride), "r" (dst1), "r" (dst2)
2153         : "%"REG_a
2154         );
2155 }
2156
2157 /**
2158  * transposes the given 8x8 block
2159  */
2160 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
2161 {
2162         asm(
2163                 "lea (%0, %1), %%"REG_a"                \n\t"
2164                 "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
2165 //      0       1       2       3       4       5       6       7       8       9
2166 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
2167                 "movq (%2), %%mm0                       \n\t" // 12345678
2168                 "movq 16(%2), %%mm1                     \n\t" // abcdefgh
2169                 "movq %%mm0, %%mm2                      \n\t" // 12345678
2170                 "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2171                 "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2172
2173                 "movq 32(%2), %%mm1                     \n\t"
2174                 "movq 48(%2), %%mm3                     \n\t"
2175                 "movq %%mm1, %%mm4                      \n\t"
2176                 "punpcklbw %%mm3, %%mm1                 \n\t"
2177                 "punpckhbw %%mm3, %%mm4                 \n\t"
2178
2179                 "movq %%mm0, %%mm3                      \n\t"
2180                 "punpcklwd %%mm1, %%mm0                 \n\t"
2181                 "punpckhwd %%mm1, %%mm3                 \n\t"
2182                 "movq %%mm2, %%mm1                      \n\t"
2183                 "punpcklwd %%mm4, %%mm2                 \n\t"
2184                 "punpckhwd %%mm4, %%mm1                 \n\t"
2185
2186                 "movd %%mm0, (%0)                       \n\t"
2187                 "psrlq $32, %%mm0                       \n\t"
2188                 "movd %%mm0, (%%"REG_a")                \n\t"
2189                 "movd %%mm3, (%%"REG_a", %1)            \n\t"
2190                 "psrlq $32, %%mm3                       \n\t"
2191                 "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
2192                 "movd %%mm2, (%0, %1, 4)                \n\t"
2193                 "psrlq $32, %%mm2                       \n\t"
2194                 "movd %%mm2, (%%"REG_d")                \n\t"
2195                 "movd %%mm1, (%%"REG_d", %1)            \n\t"
2196                 "psrlq $32, %%mm1                       \n\t"
2197                 "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
2198
2199
2200                 "movq 64(%2), %%mm0                     \n\t" // 12345678
2201                 "movq 80(%2), %%mm1                     \n\t" // abcdefgh
2202                 "movq %%mm0, %%mm2                      \n\t" // 12345678
2203                 "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2204                 "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2205
2206                 "movq 96(%2), %%mm1                     \n\t"
2207                 "movq 112(%2), %%mm3                    \n\t"
2208                 "movq %%mm1, %%mm4                      \n\t"
2209                 "punpcklbw %%mm3, %%mm1                 \n\t"
2210                 "punpckhbw %%mm3, %%mm4                 \n\t"
2211
2212                 "movq %%mm0, %%mm3                      \n\t"
2213                 "punpcklwd %%mm1, %%mm0                 \n\t"
2214                 "punpckhwd %%mm1, %%mm3                 \n\t"
2215                 "movq %%mm2, %%mm1                      \n\t"
2216                 "punpcklwd %%mm4, %%mm2                 \n\t"
2217                 "punpckhwd %%mm4, %%mm1                 \n\t"
2218
2219                 "movd %%mm0, 4(%0)                      \n\t"
2220                 "psrlq $32, %%mm0                       \n\t"
2221                 "movd %%mm0, 4(%%"REG_a")               \n\t"
2222                 "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
2223                 "psrlq $32, %%mm3                       \n\t"
2224                 "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
2225                 "movd %%mm2, 4(%0, %1, 4)               \n\t"
2226                 "psrlq $32, %%mm2                       \n\t"
2227                 "movd %%mm2, 4(%%"REG_d")               \n\t"
2228                 "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
2229                 "psrlq $32, %%mm1                       \n\t"
2230                 "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
2231
2232         :: "r" (dst), "r" ((long)dstStride), "r" (src)
2233         : "%"REG_a, "%"REG_d
2234         );
2235 }
2236 #endif //HAVE_MMX
2237 //static long test=0;
2238
2239 #ifndef HAVE_ALTIVEC
2240 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
2241                                     uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
2242 {
2243         // to save a register (FIXME do this outside of the loops)
2244         tempBluredPast[127]= maxNoise[0];
2245         tempBluredPast[128]= maxNoise[1];
2246         tempBluredPast[129]= maxNoise[2];
2247
2248 #define FAST_L2_DIFF
2249 //#define L1_DIFF //u should change the thresholds too if u try that one
2250 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2251         asm volatile(
2252                 "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
2253                 "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
2254                 "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2255 //      0       1       2       3       4       5       6       7       8       9
2256 //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
2257 //FIXME reorder?
2258 #ifdef L1_DIFF //needs mmx2
2259                 "movq (%0), %%mm0                       \n\t" // L0
2260                 "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
2261                 "movq (%0, %2), %%mm1                   \n\t" // L1
2262                 "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
2263                 "movq (%0, %2, 2), %%mm2                \n\t" // L2
2264                 "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
2265                 "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2266                 "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
2267
2268                 "movq (%0, %2, 4), %%mm4                \n\t" // L4
2269                 "paddw %%mm1, %%mm0                     \n\t"
2270                 "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
2271                 "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2272                 "paddw %%mm2, %%mm0                     \n\t"
2273                 "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
2274                 "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2275                 "paddw %%mm3, %%mm0                     \n\t"
2276                 "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
2277                 "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2278                 "paddw %%mm4, %%mm0                     \n\t"
2279                 "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
2280                 "paddw %%mm5, %%mm6                     \n\t"
2281                 "paddw %%mm7, %%mm6                     \n\t"
2282                 "paddw %%mm6, %%mm0                     \n\t"
2283 #else //L1_DIFF
2284 #if defined (FAST_L2_DIFF)
2285                 "pcmpeqb %%mm7, %%mm7                   \n\t"
2286                 "movq "MANGLE(b80)", %%mm6              \n\t"
2287                 "pxor %%mm0, %%mm0                      \n\t"
2288 #define REAL_L2_DIFF_CORE(a, b)\
2289                 "movq " #a ", %%mm5                     \n\t"\
2290                 "movq " #b ", %%mm2                     \n\t"\
2291                 "pxor %%mm7, %%mm2                      \n\t"\
2292                 PAVGB(%%mm2, %%mm5)\
2293                 "paddb %%mm6, %%mm5                     \n\t"\
2294                 "movq %%mm5, %%mm2                      \n\t"\
2295                 "psllw $8, %%mm5                        \n\t"\
2296                 "pmaddwd %%mm5, %%mm5                   \n\t"\
2297                 "pmaddwd %%mm2, %%mm2                   \n\t"\
2298                 "paddd %%mm2, %%mm5                     \n\t"\
2299                 "psrld $14, %%mm5                       \n\t"\
2300                 "paddd %%mm5, %%mm0                     \n\t"
2301
2302 #else //defined (FAST_L2_DIFF)
2303                 "pxor %%mm7, %%mm7                      \n\t"
2304                 "pxor %%mm0, %%mm0                      \n\t"
2305 #define REAL_L2_DIFF_CORE(a, b)\
2306                 "movq " #a ", %%mm5                     \n\t"\
2307                 "movq " #b ", %%mm2                     \n\t"\
2308                 "movq %%mm5, %%mm1                      \n\t"\
2309                 "movq %%mm2, %%mm3                      \n\t"\
2310                 "punpcklbw %%mm7, %%mm5                 \n\t"\
2311                 "punpckhbw %%mm7, %%mm1                 \n\t"\
2312                 "punpcklbw %%mm7, %%mm2                 \n\t"\
2313                 "punpckhbw %%mm7, %%mm3                 \n\t"\
2314                 "psubw %%mm2, %%mm5                     \n\t"\
2315                 "psubw %%mm3, %%mm1                     \n\t"\
2316                 "pmaddwd %%mm5, %%mm5                   \n\t"\
2317                 "pmaddwd %%mm1, %%mm1                   \n\t"\
2318                 "paddd %%mm1, %%mm5                     \n\t"\
2319                 "paddd %%mm5, %%mm0                     \n\t"
2320
2321 #endif //defined (FAST_L2_DIFF)
2322
2323 #define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)
2324
2325 L2_DIFF_CORE((%0)          , (%1))
2326 L2_DIFF_CORE((%0, %2)      , (%1, %2))
2327 L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))
2328 L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))
2329 L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))
2330 L2_DIFF_CORE((%0, %%REGd)  , (%1, %%REGd))
2331 L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
2332 L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))
2333
2334 #endif //L1_DIFF
2335
2336                 "movq %%mm0, %%mm4                      \n\t"
2337                 "psrlq $32, %%mm0                       \n\t"
2338                 "paddd %%mm0, %%mm4                     \n\t"
2339                 "movd %%mm4, %%ecx                      \n\t"
2340                 "shll $2, %%ecx                         \n\t"
2341                 "mov %3, %%"REG_d"                      \n\t"
2342                 "addl -4(%%"REG_d"), %%ecx              \n\t"
2343                 "addl 4(%%"REG_d"), %%ecx               \n\t"
2344                 "addl -1024(%%"REG_d"), %%ecx           \n\t"
2345                 "addl $4, %%ecx                         \n\t"
2346                 "addl 1024(%%"REG_d"), %%ecx            \n\t"
2347                 "shrl $3, %%ecx                         \n\t"
2348                 "movl %%ecx, (%%"REG_d")                \n\t"
2349
2350 //                "mov %3, %%"REG_c"                      \n\t"
2351 //                "mov %%"REG_c", test                    \n\t"
2352 //                "jmp 4f                                 \n\t"
2353                 "cmpl 512(%%"REG_d"), %%ecx             \n\t"
2354                 " jb 2f                                 \n\t"
2355                 "cmpl 516(%%"REG_d"), %%ecx             \n\t"
2356                 " jb 1f                                 \n\t"
2357
2358                 "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2359                 "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2360                 "movq (%0), %%mm0                       \n\t" // L0
2361                 "movq (%0, %2), %%mm1                   \n\t" // L1
2362                 "movq (%0, %2, 2), %%mm2                \n\t" // L2
2363                 "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2364                 "movq (%0, %2, 4), %%mm4                \n\t" // L4
2365                 "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2366                 "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2367                 "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2368                 "movq %%mm0, (%1)                       \n\t" // L0
2369                 "movq %%mm1, (%1, %2)                   \n\t" // L1
2370                 "movq %%mm2, (%1, %2, 2)                \n\t" // L2
2371                 "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
2372                 "movq %%mm4, (%1, %2, 4)                \n\t" // L4
2373                 "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
2374                 "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
2375                 "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
2376                 "jmp 4f                                 \n\t"
2377
2378                 "1:                                     \n\t"
2379                 "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2380                 "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2381                 "movq (%0), %%mm0                       \n\t" // L0
2382                 PAVGB((%1), %%mm0)                            // L0
2383                 "movq (%0, %2), %%mm1                   \n\t" // L1
2384                 PAVGB((%1, %2), %%mm1)                        // L1
2385                 "movq (%0, %2, 2), %%mm2                \n\t" // L2
2386                 PAVGB((%1, %2, 2), %%mm2)                     // L2
2387                 "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2388                 PAVGB((%1, %%REGa), %%mm3)                    // L3
2389                 "movq (%0, %2, 4), %%mm4                \n\t" // L4
2390                 PAVGB((%1, %2, 4), %%mm4)                     // L4
2391                 "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2392                 PAVGB((%1, %%REGd), %%mm5)                    // L5
2393                 "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2394                 PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
2395                 "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2396                 PAVGB((%1, %%REGc), %%mm7)                    // L7
2397                 "movq %%mm0, (%1)                       \n\t" // R0
2398                 "movq %%mm1, (%1, %2)                   \n\t" // R1
2399                 "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2400                 "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2401                 "movq %%mm4, (%1, %2, 4)                \n\t" // R4
2402                 "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
2403                 "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
2404                 "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
2405                 "movq %%mm0, (%0)                       \n\t" // L0
2406                 "movq %%mm1, (%0, %2)                   \n\t" // L1
2407                 "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2408                 "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2409                 "movq %%mm4, (%0, %2, 4)                \n\t" // L4
2410                 "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
2411                 "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
2412                 "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
2413                 "jmp 4f                                 \n\t"
2414
2415                 "2:                                     \n\t"
2416                 "cmpl 508(%%"REG_d"), %%ecx             \n\t"
2417                 " jb 3f                                 \n\t"
2418
2419                 "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2420                 "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2421                 "movq (%0), %%mm0                       \n\t" // L0
2422                 "movq (%0, %2), %%mm1                   \n\t" // L1
2423                 "movq (%0, %2, 2), %%mm2                \n\t" // L2
2424                 "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2425                 "movq (%1), %%mm4                       \n\t" // R0
2426                 "movq (%1, %2), %%mm5                   \n\t" // R1
2427                 "movq (%1, %2, 2), %%mm6                \n\t" // R2
2428                 "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2429                 PAVGB(%%mm4, %%mm0)
2430                 PAVGB(%%mm5, %%mm1)
2431                 PAVGB(%%mm6, %%mm2)
2432                 PAVGB(%%mm7, %%mm3)
2433                 PAVGB(%%mm4, %%mm0)
2434                 PAVGB(%%mm5, %%mm1)
2435                 PAVGB(%%mm6, %%mm2)
2436                 PAVGB(%%mm7, %%mm3)
2437                 "movq %%mm0, (%1)                       \n\t" // R0
2438                 "movq %%mm1, (%1, %2)                   \n\t" // R1
2439                 "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2440                 "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2441                 "movq %%mm0, (%0)                       \n\t" // L0
2442                 "movq %%mm1, (%0, %2)                   \n\t" // L1
2443                 "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2444                 "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2445
2446                 "movq (%0, %2, 4), %%mm0                \n\t" // L4
2447                 "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2448                 "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2449                 "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2450                 "movq (%1, %2, 4), %%mm4                \n\t" // R4
2451                 "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2452                 "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2453                 "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2454                 PAVGB(%%mm4, %%mm0)
2455                 PAVGB(%%mm5, %%mm1)
2456                 PAVGB(%%mm6, %%mm2)
2457                 PAVGB(%%mm7, %%mm3)
2458                 PAVGB(%%mm4, %%mm0)
2459                 PAVGB(%%mm5, %%mm1)
2460                 PAVGB(%%mm6, %%mm2)
2461                 PAVGB(%%mm7, %%mm3)
2462                 "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2463                 "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2464                 "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2465                 "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2466                 "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2467                 "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2468                 "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2469                 "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2470                 "jmp 4f                                 \n\t"
2471
2472                 "3:                                     \n\t"
2473                 "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2474                 "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2475                 "movq (%0), %%mm0                       \n\t" // L0
2476                 "movq (%0, %2), %%mm1                   \n\t" // L1
2477                 "movq (%0, %2, 2), %%mm2                \n\t" // L2
2478                 "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2479                 "movq (%1), %%mm4                       \n\t" // R0
2480                 "movq (%1, %2), %%mm5                   \n\t" // R1
2481                 "movq (%1, %2, 2), %%mm6                \n\t" // R2
2482                 "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2483                 PAVGB(%%mm4, %%mm0)
2484                 PAVGB(%%mm5, %%mm1)
2485                 PAVGB(%%mm6, %%mm2)
2486                 PAVGB(%%mm7, %%mm3)
2487                 PAVGB(%%mm4, %%mm0)
2488                 PAVGB(%%mm5, %%mm1)
2489                 PAVGB(%%mm6, %%mm2)
2490                 PAVGB(%%mm7, %%mm3)
2491                 PAVGB(%%mm4, %%mm0)
2492                 PAVGB(%%mm5, %%mm1)
2493                 PAVGB(%%mm6, %%mm2)
2494                 PAVGB(%%mm7, %%mm3)
2495                 "movq %%mm0, (%1)                       \n\t" // R0
2496                 "movq %%mm1, (%1, %2)                   \n\t" // R1
2497                 "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2498                 "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2499                 "movq %%mm0, (%0)                       \n\t" // L0
2500                 "movq %%mm1, (%0, %2)                   \n\t" // L1
2501                 "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2502                 "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2503
2504                 "movq (%0, %2, 4), %%mm0                \n\t" // L4
2505                 "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2506                 "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2507                 "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2508                 "movq (%1, %2, 4), %%mm4                \n\t" // R4
2509                 "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2510                 "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2511                 "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2512                 PAVGB(%%mm4, %%mm0)
2513                 PAVGB(%%mm5, %%mm1)
2514                 PAVGB(%%mm6, %%mm2)
2515                 PAVGB(%%mm7, %%mm3)
2516                 PAVGB(%%mm4, %%mm0)
2517                 PAVGB(%%mm5, %%mm1)
2518                 PAVGB(%%mm6, %%mm2)
2519                 PAVGB(%%mm7, %%mm3)
2520                 PAVGB(%%mm4, %%mm0)
2521                 PAVGB(%%mm5, %%mm1)
2522                 PAVGB(%%mm6, %%mm2)
2523                 PAVGB(%%mm7, %%mm3)
2524                 "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2525                 "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2526                 "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2527                 "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2528                 "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2529                 "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2530                 "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2531                 "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2532
2533                 "4:                                     \n\t"
2534
2535                 :: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
2536                 : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
2537                 );
2538 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2539 {
2540         int y;
2541         int d=0;
2542 //        int sysd=0;
2543         int i;
2544
2545         for(y=0; y<8; y++)
2546         {
2547                 int x;
2548                 for(x=0; x<8; x++)
2549                 {
2550                         int ref= tempBlured[ x + y*stride ];
2551                         int cur= src[ x + y*stride ];
2552                         int d1=ref - cur;
2553 //                        if(x==0 || x==7) d1+= d1>>1;
2554 //                        if(y==0 || y==7) d1+= d1>>1;
2555 //                        d+= FFABS(d1);
2556                         d+= d1*d1;
2557 //                        sysd+= d1;
2558                 }
2559         }
2560         i=d;
2561         d=         (
2562                 4*d
2563                 +(*(tempBluredPast-256))
2564                 +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
2565                 +(*(tempBluredPast+256))
2566                 +4)>>3;
2567         *tempBluredPast=i;
2568 //        ((*tempBluredPast)*3 + d + 2)>>2;
2569
2570 /*
2571 Switch between
2572  1  0  0  0  0  0  0  (0)
2573 64 32 16  8  4  2  1  (1)
2574 64 48 36 27 20 15 11 (33) (approx)
2575 64 56 49 43 37 33 29 (200) (approx)
2576 */
2577         if(d > maxNoise[1])
2578         {
2579                 if(d < maxNoise[2])
2580                 {
2581                         for(y=0; y<8; y++)
2582                         {
2583                                 int x;
2584                                 for(x=0; x<8; x++)
2585                                 {
2586                                         int ref= tempBlured[ x + y*stride ];
2587                                         int cur= src[ x + y*stride ];
2588                                         tempBlured[ x + y*stride ]=
2589                                         src[ x + y*stride ]=
2590                                                 (ref + cur + 1)>>1;
2591                                 }
2592                         }
2593                 }
2594                 else
2595                 {
2596                         for(y=0; y<8; y++)
2597                         {
2598                                 int x;
2599                                 for(x=0; x<8; x++)
2600                                 {
2601                                         tempBlured[ x + y*stride ]= src[ x + y*stride ];
2602                                 }
2603                         }
2604                 }
2605         }
2606         else
2607         {
2608                 if(d < maxNoise[0])
2609                 {
2610                         for(y=0; y<8; y++)
2611                         {
2612                                 int x;
2613                                 for(x=0; x<8; x++)
2614                                 {
2615                                         int ref= tempBlured[ x + y*stride ];
2616                                         int cur= src[ x + y*stride ];
2617                                         tempBlured[ x + y*stride ]=
2618                                         src[ x + y*stride ]=
2619                                                 (ref*7 + cur + 4)>>3;
2620                                 }
2621                         }
2622                 }
2623                 else
2624                 {
2625                         for(y=0; y<8; y++)
2626                         {
2627                                 int x;
2628                                 for(x=0; x<8; x++)
2629                                 {
2630                                         int ref= tempBlured[ x + y*stride ];
2631                                         int cur= src[ x + y*stride ];
2632                                         tempBlured[ x + y*stride ]=
2633                                         src[ x + y*stride ]=
2634                                                 (ref*3 + cur + 2)>>2;
2635                                 }
2636                         }
2637                 }
2638         }
2639 }
2640 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2641 }
2642 #endif //HAVE_ALTIVEC
2643
2644 #ifdef HAVE_MMX
2645 /**
2646  * accurate deblock filter
2647  */
2648 static always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
2649         int64_t dc_mask, eq_mask, both_masks;
2650         int64_t sums[10*8*2];
2651         src+= step*3; // src points to begin of the 8x8 Block
2652 //START_TIMER
2653 asm volatile(
2654                 "movq %0, %%mm7                         \n\t"
2655                 "movq %1, %%mm6                         \n\t"
2656                 : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
2657                 );
2658
2659 asm volatile(
2660                 "lea (%2, %3), %%"REG_a"                \n\t"
2661 //      0       1       2       3       4       5       6       7       8       9
2662 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
2663
2664                 "movq (%2), %%mm0                       \n\t"
2665                 "movq (%%"REG_a"), %%mm1                \n\t"
2666                 "movq %%mm1, %%mm3                      \n\t"
2667                 "movq %%mm1, %%mm4                      \n\t"
2668                 "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
2669                 "paddb %%mm7, %%mm0                     \n\t"
2670                 "pcmpgtb %%mm6, %%mm0                   \n\t"
2671
2672                 "movq (%%"REG_a",%3), %%mm2             \n\t"
2673                 PMAXUB(%%mm2, %%mm4)
2674                 PMINUB(%%mm2, %%mm3, %%mm5)
2675                 "psubb %%mm2, %%mm1                     \n\t"
2676                 "paddb %%mm7, %%mm1                     \n\t"
2677                 "pcmpgtb %%mm6, %%mm1                   \n\t"
2678                 "paddb %%mm1, %%mm0                     \n\t"
2679
2680                 "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2681                 PMAXUB(%%mm1, %%mm4)
2682                 PMINUB(%%mm1, %%mm3, %%mm5)
2683                 "psubb %%mm1, %%mm2                     \n\t"
2684                 "paddb %%mm7, %%mm2                     \n\t"
2685                 "pcmpgtb %%mm6, %%mm2                   \n\t"
2686                 "paddb %%mm2, %%mm0                     \n\t"
2687
2688                 "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
2689
2690                 "movq (%2, %3, 4), %%mm2                \n\t"
2691                 PMAXUB(%%mm2, %%mm4)
2692                 PMINUB(%%mm2, %%mm3, %%mm5)
2693                 "psubb %%mm2, %%mm1                     \n\t"
2694                 "paddb %%mm7, %%mm1                     \n\t"
2695                 "pcmpgtb %%mm6, %%mm1                   \n\t"
2696                 "paddb %%mm1, %%mm0                     \n\t"
2697
2698                 "movq (%%"REG_a"), %%mm1                \n\t"
2699                 PMAXUB(%%mm1, %%mm4)
2700                 PMINUB(%%mm1, %%mm3, %%mm5)
2701                 "psubb %%mm1, %%mm2                     \n\t"
2702                 "paddb %%mm7, %%mm2                     \n\t"
2703                 "pcmpgtb %%mm6, %%mm2                   \n\t"
2704                 "paddb %%mm2, %%mm0                     \n\t"
2705
2706                 "movq (%%"REG_a", %3), %%mm2            \n\t"
2707                 PMAXUB(%%mm2, %%mm4)
2708                 PMINUB(%%mm2, %%mm3, %%mm5)
2709                 "psubb %%mm2, %%mm1                     \n\t"
2710                 "paddb %%mm7, %%mm1                     \n\t"
2711                 "pcmpgtb %%mm6, %%mm1                   \n\t"
2712                 "paddb %%mm1, %%mm0                     \n\t"
2713
2714                 "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2715                 PMAXUB(%%mm1, %%mm4)
2716                 PMINUB(%%mm1, %%mm3, %%mm5)
2717                 "psubb %%mm1, %%mm2                     \n\t"
2718                 "paddb %%mm7, %%mm2                     \n\t"
2719                 "pcmpgtb %%mm6, %%mm2                   \n\t"
2720                 "paddb %%mm2, %%mm0                     \n\t"
2721
2722                 "movq (%2, %3, 8), %%mm2                \n\t"
2723                 PMAXUB(%%mm2, %%mm4)
2724                 PMINUB(%%mm2, %%mm3, %%mm5)
2725                 "psubb %%mm2, %%mm1                     \n\t"
2726                 "paddb %%mm7, %%mm1                     \n\t"
2727                 "pcmpgtb %%mm6, %%mm1                   \n\t"
2728                 "paddb %%mm1, %%mm0                     \n\t"
2729
2730                 "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
2731                 "psubb %%mm1, %%mm2                     \n\t"
2732                 "paddb %%mm7, %%mm2                     \n\t"
2733                 "pcmpgtb %%mm6, %%mm2                   \n\t"
2734                 "paddb %%mm2, %%mm0                     \n\t"
2735                 "psubusb %%mm3, %%mm4                   \n\t"
2736
2737                 "pxor %%mm6, %%mm6                      \n\t"
2738                 "movq %4, %%mm7                         \n\t" // QP,..., QP
2739                 "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
2740                 "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
2741                 "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2742                 "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2743                 "movq %%mm7, %1                         \n\t"
2744
2745                 "movq %5, %%mm7                         \n\t"
2746                 "punpcklbw %%mm7, %%mm7                 \n\t"
2747                 "punpcklbw %%mm7, %%mm7                 \n\t"
2748                 "punpcklbw %%mm7, %%mm7                 \n\t"
2749                 "psubb %%mm0, %%mm6                     \n\t"
2750                 "pcmpgtb %%mm7, %%mm6                   \n\t"
2751                 "movq %%mm6, %0                         \n\t"
2752
2753                 : "=m" (eq_mask), "=m" (dc_mask)
2754                 : "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2755                 : "%"REG_a
2756                 );
2757
2758         both_masks = dc_mask & eq_mask;
2759
2760         if(both_masks){
2761                 long offset= -8*step;
2762                 int64_t *temp_sums= sums;
2763
2764                 asm volatile(
2765                 "movq %2, %%mm0                         \n\t"  // QP,..., QP
2766                 "pxor %%mm4, %%mm4                      \n\t"
2767
2768                 "movq (%0), %%mm6                       \n\t"
2769                 "movq (%0, %1), %%mm5                   \n\t"
2770                 "movq %%mm5, %%mm1                      \n\t"
2771                 "movq %%mm6, %%mm2                      \n\t"
2772                 "psubusb %%mm6, %%mm5                   \n\t"
2773                 "psubusb %%mm1, %%mm2                   \n\t"
2774                 "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2775                 "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2776                 "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2777
2778                 "pxor %%mm6, %%mm1                      \n\t"
2779                 "pand %%mm0, %%mm1                      \n\t"
2780                 "pxor %%mm1, %%mm6                      \n\t"
2781                 // 0:QP  6:First
2782
2783                 "movq (%0, %1, 8), %%mm5                \n\t"
2784                 "add %1, %0                             \n\t" // %0 points to line 1 not 0
2785                 "movq (%0, %1, 8), %%mm7                \n\t"
2786                 "movq %%mm5, %%mm1                      \n\t"
2787                 "movq %%mm7, %%mm2                      \n\t"
2788                 "psubusb %%mm7, %%mm5                   \n\t"
2789                 "psubusb %%mm1, %%mm2                   \n\t"
2790                 "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2791                 "movq %2, %%mm0                         \n\t"  // QP,..., QP
2792                 "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2793                 "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2794
2795                 "pxor %%mm7, %%mm1                      \n\t"
2796                 "pand %%mm0, %%mm1                      \n\t"
2797                 "pxor %%mm1, %%mm7                      \n\t"
2798
2799                 "movq %%mm6, %%mm5                      \n\t"
2800                 "punpckhbw %%mm4, %%mm6                 \n\t"
2801                 "punpcklbw %%mm4, %%mm5                 \n\t"
2802                 // 4:0 5/6:First 7:Last
2803
2804                 "movq %%mm5, %%mm0                      \n\t"
2805                 "movq %%mm6, %%mm1                      \n\t"
2806                 "psllw $2, %%mm0                        \n\t"
2807                 "psllw $2, %%mm1                        \n\t"
2808                 "paddw "MANGLE(w04)", %%mm0             \n\t"
2809                 "paddw "MANGLE(w04)", %%mm1             \n\t"
2810
2811 #define NEXT\
2812                 "movq (%0), %%mm2                       \n\t"\
2813                 "movq (%0), %%mm3                       \n\t"\
2814                 "add %1, %0                             \n\t"\
2815                 "punpcklbw %%mm4, %%mm2                 \n\t"\
2816                 "punpckhbw %%mm4, %%mm3                 \n\t"\
2817                 "paddw %%mm2, %%mm0                     \n\t"\
2818                 "paddw %%mm3, %%mm1                     \n\t"
2819
2820 #define PREV\
2821                 "movq (%0), %%mm2                       \n\t"\
2822                 "movq (%0), %%mm3                       \n\t"\
2823                 "add %1, %0                             \n\t"\
2824                 "punpcklbw %%mm4, %%mm2                 \n\t"\
2825                 "punpckhbw %%mm4, %%mm3                 \n\t"\
2826                 "psubw %%mm2, %%mm0                     \n\t"\
2827                 "psubw %%mm3, %%mm1                     \n\t"
2828
2829
2830                 NEXT //0
2831                 NEXT //1
2832                 NEXT //2
2833                 "movq %%mm0, (%3)                       \n\t"
2834                 "movq %%mm1, 8(%3)                      \n\t"
2835
2836                 NEXT //3
2837                 "psubw %%mm5, %%mm0                     \n\t"
2838                 "psubw %%mm6, %%mm1                     \n\t"
2839                 "movq %%mm0, 16(%3)                     \n\t"
2840                 "movq %%mm1, 24(%3)                     \n\t"
2841
2842                 NEXT //4
2843                 "psubw %%mm5, %%mm0                     \n\t"
2844                 "psubw %%mm6, %%mm1                     \n\t"
2845                 "movq %%mm0, 32(%3)                     \n\t"
2846                 "movq %%mm1, 40(%3)                     \n\t"
2847
2848                 NEXT //5
2849                 "psubw %%mm5, %%mm0                     \n\t"
2850                 "psubw %%mm6, %%mm1                     \n\t"
2851                 "movq %%mm0, 48(%3)                     \n\t"
2852                 "movq %%mm1, 56(%3)                     \n\t"
2853
2854                 NEXT //6
2855                 "psubw %%mm5, %%mm0                     \n\t"
2856                 "psubw %%mm6, %%mm1                     \n\t"
2857                 "movq %%mm0, 64(%3)                     \n\t"
2858                 "movq %%mm1, 72(%3)                     \n\t"
2859
2860                 "movq %%mm7, %%mm6                      \n\t"
2861                 "punpckhbw %%mm4, %%mm7                 \n\t"
2862                 "punpcklbw %%mm4, %%mm6                 \n\t"
2863
2864                 NEXT //7
2865                 "mov %4, %0                             \n\t"
2866                 "add %1, %0                             \n\t"
2867                 PREV //0
2868                 "movq %%mm0, 80(%3)                     \n\t"
2869                 "movq %%mm1, 88(%3)                     \n\t"
2870
2871                 PREV //1
2872                 "paddw %%mm6, %%mm0                     \n\t"
2873                 "paddw %%mm7, %%mm1                     \n\t"
2874                 "movq %%mm0, 96(%3)                     \n\t"
2875                 "movq %%mm1, 104(%3)                    \n\t"
2876
2877                 PREV //2
2878                 "paddw %%mm6, %%mm0                     \n\t"
2879                 "paddw %%mm7, %%mm1                     \n\t"
2880                 "movq %%mm0, 112(%3)                    \n\t"
2881                 "movq %%mm1, 120(%3)                    \n\t"
2882
2883                 PREV //3
2884                 "paddw %%mm6, %%mm0                     \n\t"
2885                 "paddw %%mm7, %%mm1                     \n\t"
2886                 "movq %%mm0, 128(%3)                    \n\t"
2887                 "movq %%mm1, 136(%3)                    \n\t"
2888
2889                 PREV //4
2890                 "paddw %%mm6, %%mm0                     \n\t"
2891                 "paddw %%mm7, %%mm1                     \n\t"
2892                 "movq %%mm0, 144(%3)                    \n\t"
2893                 "movq %%mm1, 152(%3)                    \n\t"
2894
2895                 "mov %4, %0                             \n\t" //FIXME
2896
2897                 : "+&r"(src)
2898                 : "r" ((long)step), "m" (c->pQPb), "r"(sums), "g"(src)
2899                 );
2900
2901                 src+= step; // src points to begin of the 8x8 Block
2902
2903                 asm volatile(
2904                 "movq %4, %%mm6                         \n\t"
2905                 "pcmpeqb %%mm5, %%mm5                   \n\t"
2906                 "pxor %%mm6, %%mm5                      \n\t"
2907                 "pxor %%mm7, %%mm7                      \n\t"
2908
2909                 "1:                                     \n\t"
2910                 "movq (%1), %%mm0                       \n\t"
2911                 "movq 8(%1), %%mm1                      \n\t"
2912                 "paddw 32(%1), %%mm0                    \n\t"
2913                 "paddw 40(%1), %%mm1                    \n\t"
2914                 "movq (%0, %3), %%mm2                   \n\t"
2915                 "movq %%mm2, %%mm3                      \n\t"
2916                 "movq %%mm2, %%mm4                      \n\t"
2917                 "punpcklbw %%mm7, %%mm2                 \n\t"
2918                 "punpckhbw %%mm7, %%mm3                 \n\t"
2919                 "paddw %%mm2, %%mm0                     \n\t"
2920                 "paddw %%mm3, %%mm1                     \n\t"
2921                 "paddw %%mm2, %%mm0                     \n\t"
2922                 "paddw %%mm3, %%mm1                     \n\t"
2923                 "psrlw $4, %%mm0                        \n\t"
2924                 "psrlw $4, %%mm1                        \n\t"
2925                 "packuswb %%mm1, %%mm0                  \n\t"
2926                 "pand %%mm6, %%mm0                      \n\t"
2927                 "pand %%mm5, %%mm4                      \n\t"
2928                 "por %%mm4, %%mm0                       \n\t"
2929                 "movq %%mm0, (%0, %3)                   \n\t"
2930                 "add $16, %1                            \n\t"
2931                 "add %2, %0                             \n\t"
2932                 " js 1b                                 \n\t"
2933
2934                 : "+r"(offset), "+r"(temp_sums)
2935                 : "r" ((long)step), "r"(src - offset), "m"(both_masks)
2936                 );
2937         }else
2938                 src+= step; // src points to begin of the 8x8 Block
2939
2940         if(eq_mask != -1LL){
2941                 uint8_t *temp_src= src;
2942                 asm volatile(
2943                 "pxor %%mm7, %%mm7                      \n\t"
2944                 "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
2945                 "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
2946 //      0       1       2       3       4       5       6       7       8       9
2947 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1
2948
2949                 "movq (%0), %%mm0                       \n\t"
2950                 "movq %%mm0, %%mm1                      \n\t"
2951                 "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
2952                 "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
2953
2954                 "movq (%0, %1), %%mm2                   \n\t"
2955                 "lea (%0, %1, 2), %%"REG_a"             \n\t"
2956                 "movq %%mm2, %%mm3                      \n\t"
2957                 "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
2958                 "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
2959
2960                 "movq (%%"REG_a"), %%mm4                \n\t"
2961                 "movq %%mm4, %%mm5                      \n\t"
2962                 "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
2963                 "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
2964
2965                 "paddw %%mm0, %%mm0                     \n\t" // 2L0
2966                 "paddw %%mm1, %%mm1                     \n\t" // 2H0
2967                 "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
2968                 "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
2969                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
2970                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
2971
2972                 "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
2973                 "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
2974                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
2975                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
2976
2977                 "movq (%%"REG_a", %1), %%mm2            \n\t"
2978                 "movq %%mm2, %%mm3                      \n\t"
2979                 "punpcklbw %%mm7, %%mm2                 \n\t" // L3
2980                 "punpckhbw %%mm7, %%mm3                 \n\t" // H3
2981
2982                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
2983                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
2984                 "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2985                 "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2986                 "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2987                 "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2988
2989                 "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
2990                 "movq %%mm0, %%mm1                      \n\t"
2991                 "punpcklbw %%mm7, %%mm0                 \n\t" // L4
2992                 "punpckhbw %%mm7, %%mm1                 \n\t" // H4
2993
2994                 "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
2995                 "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
2996                 "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
2997                 "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
2998                 "paddw %%mm4, %%mm4                     \n\t" // 2L2
2999                 "paddw %%mm5, %%mm5                     \n\t" // 2H2
3000                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
3001                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
3002
3003                 "lea (%%"REG_a", %1), %0                \n\t"
3004                 "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
3005                 "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
3006                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
3007                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
3008 //50 opcodes so far
3009                 "movq (%0, %1, 2), %%mm2                \n\t"
3010                 "movq %%mm2, %%mm3                      \n\t"
3011                 "punpcklbw %%mm7, %%mm2                 \n\t" // L5
3012                 "punpckhbw %%mm7, %%mm3                 \n\t" // H5
3013                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
3014                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
3015                 "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
3016                 "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
3017
3018                 "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
3019                 "punpcklbw %%mm7, %%mm6                 \n\t" // L6
3020                 "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
3021                 "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
3022                 "punpckhbw %%mm7, %%mm6                 \n\t" // H6
3023                 "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
3024
3025                 "paddw %%mm0, %%mm0                     \n\t" // 2L4
3026                 "paddw %%mm1, %%mm1                     \n\t" // 2H4
3027                 "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
3028                 "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
3029
3030                 "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
3031                 "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
3032                 "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
3033                 "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
3034
3035                 "movq (%0, %1, 4), %%mm2                \n\t"
3036                 "movq %%mm2, %%mm3                      \n\t"
3037                 "punpcklbw %%mm7, %%mm2                 \n\t" // L7
3038                 "punpckhbw %%mm7, %%mm3                 \n\t" // H7
3039
3040                 "paddw %%mm2, %%mm2                     \n\t" // 2L7
3041                 "paddw %%mm3, %%mm3                     \n\t" // 2H7
3042                 "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
3043                 "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
3044
3045                 "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
3046                 "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
3047
3048 #ifdef HAVE_MMX2
3049                 "movq %%mm7, %%mm6                      \n\t" // 0
3050                 "psubw %%mm0, %%mm6                     \n\t"
3051                 "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3052                 "movq %%mm7, %%mm6                      \n\t" // 0
3053                 "psubw %%mm1, %%mm6                     \n\t"
3054                 "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3055                 "movq %%mm7, %%mm6                      \n\t" // 0
3056                 "psubw %%mm2, %%mm6                     \n\t"
3057                 "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3058                 "movq %%mm7, %%mm6                      \n\t" // 0
3059                 "psubw %%mm3, %%mm6                     \n\t"
3060                 "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3061 #else
3062                 "movq %%mm7, %%mm6                      \n\t" // 0
3063                 "pcmpgtw %%mm0, %%mm6                   \n\t"
3064                 "pxor %%mm6, %%mm0                      \n\t"
3065                 "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3066                 "movq %%mm7, %%mm6                      \n\t" // 0
3067                 "pcmpgtw %%mm1, %%mm6                   \n\t"
3068                 "pxor %%mm6, %%mm1                      \n\t"
3069                 "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3070                 "movq %%mm7, %%mm6                      \n\t" // 0
3071                 "pcmpgtw %%mm2, %%mm6                   \n\t"
3072                 "pxor %%mm6, %%mm2                      \n\t"
3073                 "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3074                 "movq %%mm7, %%mm6                      \n\t" // 0
3075                 "pcmpgtw %%mm3, %%mm6                   \n\t"
3076                 "pxor %%mm6, %%mm3                      \n\t"
3077                 "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3078 #endif
3079
3080 #ifdef HAVE_MMX2
3081                 "pminsw %%mm2, %%mm0                    \n\t"
3082                 "pminsw %%mm3, %%mm1                    \n\t"
3083 #else
3084                 "movq %%mm0, %%mm6                      \n\t"
3085                 "psubusw %%mm2, %%mm6                   \n\t"
3086                 "psubw %%mm6, %%mm0                     \n\t"
3087                 "movq %%mm1, %%mm6                      \n\t"
3088                 "psubusw %%mm3, %%mm6                   \n\t"
3089                 "psubw %%mm6, %%mm1                     \n\t"
3090 #endif
3091
3092                 "movd %2, %%mm2                         \n\t" // QP
3093                 "punpcklbw %%mm7, %%mm2                 \n\t"
3094
3095                 "movq %%mm7, %%mm6                      \n\t" // 0
3096                 "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
3097                 "pxor %%mm6, %%mm4                      \n\t"
3098                 "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
3099                 "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
3100                 "pxor %%mm7, %%mm5                      \n\t"
3101                 "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
3102 // 100 opcodes
3103                 "psllw $3, %%mm2                        \n\t" // 8QP
3104                 "movq %%mm2, %%mm3                      \n\t" // 8QP
3105                 "pcmpgtw %%mm4, %%mm2                   \n\t"
3106                 "pcmpgtw %%mm5, %%mm3                   \n\t"
3107                 "pand %%mm2, %%mm4                      \n\t"
3108                 "pand %%mm3, %%mm5                      \n\t"
3109
3110
3111                 "psubusw %%mm0, %%mm4                   \n\t" // hd
3112                 "psubusw %%mm1, %%mm5                   \n\t" // ld
3113
3114
3115                 "movq "MANGLE(w05)", %%mm2              \n\t" // 5
3116                 "pmullw %%mm2, %%mm4                    \n\t"
3117                 "pmullw %%mm2, %%mm5                    \n\t"
3118                 "movq "MANGLE(w20)", %%mm2              \n\t" // 32
3119                 "paddw %%mm2, %%mm4                     \n\t"
3120                 "paddw %%mm2, %%mm5                     \n\t"
3121                 "psrlw $6, %%mm4                        \n\t"
3122                 "psrlw $6, %%mm5                        \n\t"
3123
3124                 "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
3125                 "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
3126
3127                 "pxor %%mm2, %%mm2                      \n\t"
3128                 "pxor %%mm3, %%mm3                      \n\t"
3129
3130                 "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
3131                 "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
3132                 "pxor %%mm2, %%mm0                      \n\t"
3133                 "pxor %%mm3, %%mm1                      \n\t"
3134                 "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
3135                 "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
3136                 "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
3137                 "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
3138
3139                 "pxor %%mm6, %%mm2                      \n\t"
3140                 "pxor %%mm7, %%mm3                      \n\t"
3141                 "pand %%mm2, %%mm4                      \n\t"
3142                 "pand %%mm3, %%mm5                      \n\t"
3143
3144 #ifdef HAVE_MMX2
3145                 "pminsw %%mm0, %%mm4                    \n\t"
3146                 "pminsw %%mm1, %%mm5                    \n\t"
3147 #else
3148                 "movq %%mm4, %%mm2                      \n\t"
3149                 "psubusw %%mm0, %%mm2                   \n\t"
3150                 "psubw %%mm2, %%mm4                     \n\t"
3151                 "movq %%mm5, %%mm2                      \n\t"
3152                 "psubusw %%mm1, %%mm2                   \n\t"
3153                 "psubw %%mm2, %%mm5                     \n\t"
3154 #endif
3155                 "pxor %%mm6, %%mm4                      \n\t"
3156                 "pxor %%mm7, %%mm5                      \n\t"
3157                 "psubw %%mm6, %%mm4                     \n\t"
3158                 "psubw %%mm7, %%mm5                     \n\t"
3159                 "packsswb %%mm5, %%mm4                  \n\t"
3160                 "movq %3, %%mm1                         \n\t"
3161                 "pandn %%mm4, %%mm1                     \n\t"
3162                 "movq (%0), %%mm0                       \n\t"
3163                 "paddb   %%mm1, %%mm0                   \n\t"
3164                 "movq %%mm0, (%0)                       \n\t"
3165                 "movq (%0, %1), %%mm0                   \n\t"
3166                 "psubb %%mm1, %%mm0                     \n\t"
3167                 "movq %%mm0, (%0, %1)                   \n\t"
3168
3169                 : "+r" (temp_src)
3170                 : "r" ((long)step), "m" (c->pQPb), "m"(eq_mask)
3171                 : "%"REG_a, "%"REG_c
3172                 );
3173         }
3174 /*if(step==16){
3175     STOP_TIMER("step16")
3176 }else{
3177     STOP_TIMER("stepX")
3178 }*/
3179 }
3180 #endif //HAVE_MMX
3181
3182 static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3183         QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
3184
3185 /**
3186  * Copies a block from src to dst and fixes the blacklevel
3187  * levelFix == 0 -> dont touch the brighness & contrast
3188  */
3189 #undef SCALED_CPY
3190
3191 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
3192         int levelFix, int64_t *packedOffsetAndScale)
3193 {
3194 #ifndef HAVE_MMX
3195         int i;
3196 #endif
3197         if(levelFix)
3198         {
3199 #ifdef HAVE_MMX
3200                 asm volatile(
3201                         "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
3202                         "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
3203                         "lea (%2,%4), %%"REG_a"         \n\t"
3204                         "lea (%3,%5), %%"REG_d"         \n\t"
3205                         "pxor %%mm4, %%mm4              \n\t"
3206 #ifdef HAVE_MMX2
3207 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \
3208                         "movq " #src1 ", %%mm0          \n\t"\
3209                         "movq " #src1 ", %%mm5          \n\t"\
3210                         "movq " #src2 ", %%mm1          \n\t"\
3211                         "movq " #src2 ", %%mm6          \n\t"\
3212                         "punpcklbw %%mm0, %%mm0         \n\t"\
3213                         "punpckhbw %%mm5, %%mm5         \n\t"\
3214                         "punpcklbw %%mm1, %%mm1         \n\t"\
3215                         "punpckhbw %%mm6, %%mm6         \n\t"\
3216                         "pmulhuw %%mm3, %%mm0           \n\t"\
3217                         "pmulhuw %%mm3, %%mm5           \n\t"\
3218                         "pmulhuw %%mm3, %%mm1           \n\t"\
3219                         "pmulhuw %%mm3, %%mm6           \n\t"\
3220                         "psubw %%mm2, %%mm0             \n\t"\
3221                         "psubw %%mm2, %%mm5             \n\t"\
3222                         "psubw %%mm2, %%mm1             \n\t"\
3223                         "psubw %%mm2, %%mm6             \n\t"\
3224                         "packuswb %%mm5, %%mm0          \n\t"\
3225                         "packuswb %%mm6, %%mm1          \n\t"\
3226                         "movq %%mm0, " #dst1 "          \n\t"\
3227                         "movq %%mm1, " #dst2 "          \n\t"\
3228
3229 #else //HAVE_MMX2
3230 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \
3231                         "movq " #src1 ", %%mm0          \n\t"\
3232                         "movq " #src1 ", %%mm5          \n\t"\
3233                         "punpcklbw %%mm4, %%mm0         \n\t"\
3234                         "punpckhbw %%mm4, %%mm5         \n\t"\
3235                         "psubw %%mm2, %%mm0             \n\t"\
3236                         "psubw %%mm2, %%mm5             \n\t"\
3237                         "movq " #src2 ", %%mm1          \n\t"\
3238                         "psllw $6, %%mm0                \n\t"\
3239                         "psllw $6, %%mm5                \n\t"\
3240                         "pmulhw %%mm3, %%mm0            \n\t"\
3241                         "movq " #src2 ", %%mm6          \n\t"\
3242                         "pmulhw %%mm3, %%mm5            \n\t"\
3243                         "punpcklbw %%mm4, %%mm1         \n\t"\
3244                         "punpckhbw %%mm4, %%mm6         \n\t"\
3245                         "psubw %%mm2, %%mm1             \n\t"\
3246                         "psubw %%mm2, %%mm6             \n\t"\
3247                         "psllw $6, %%mm1                \n\t"\
3248                         "psllw $6, %%mm6                \n\t"\
3249                         "pmulhw %%mm3, %%mm1            \n\t"\
3250                         "pmulhw %%mm3, %%mm6            \n\t"\
3251                         "packuswb %%mm5, %%mm0          \n\t"\
3252                         "packuswb %%mm6, %%mm1          \n\t"\
3253                         "movq %%mm0, " #dst1 "          \n\t"\
3254                         "movq %%mm1, " #dst2 "          \n\t"\
3255
3256 #endif //HAVE_MMX2
3257 #define SCALED_CPY(src1, src2, dst1, dst2)\
3258    REAL_SCALED_CPY(src1, src2, dst1, dst2)
3259
3260 SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
3261 SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
3262 SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
3263                         "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
3264                         "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
3265 SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
3266
3267
3268                         : "=&a" (packedOffsetAndScale)
3269                         : "0" (packedOffsetAndScale),
3270                         "r"(src),
3271                         "r"(dst),
3272                         "r" ((long)srcStride),
3273                         "r" ((long)dstStride)
3274                         : "%"REG_d
3275                                         );
3276 #else //HAVE_MMX
3277         for(i=0; i<8; i++)
3278                 memcpy( &(dst[dstStride*i]),
3279                         &(src[srcStride*i]), BLOCK_SIZE);
3280 #endif //HAVE_MMX
3281         }
3282         else
3283         {
3284 #ifdef HAVE_MMX
3285         asm volatile(
3286                 "lea (%0,%2), %%"REG_a"                 \n\t"
3287                 "lea (%1,%3), %%"REG_d"                 \n\t"
3288
3289 #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \
3290                 "movq " #src1 ", %%mm0          \n\t"\
3291                 "movq " #src2 ", %%mm1          \n\t"\
3292                 "movq %%mm0, " #dst1 "          \n\t"\
3293                 "movq %%mm1, " #dst2 "          \n\t"\
3294
3295 #define SIMPLE_CPY(src1, src2, dst1, dst2)\
3296    REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
3297
3298 SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))
3299 SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
3300 SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
3301                 "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
3302                 "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
3303 SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
3304
3305                 : : "r" (src),
3306                 "r" (dst),
3307                 "r" ((long)srcStride),
3308                 "r" ((long)dstStride)
3309                 : "%"REG_a, "%"REG_d
3310         );
3311 #else //HAVE_MMX
3312         for(i=0; i<8; i++)
3313                 memcpy( &(dst[dstStride*i]),
3314                         &(src[srcStride*i]), BLOCK_SIZE);
3315 #endif //HAVE_MMX
3316         }
3317 }
3318
3319 /**
3320  * Duplicates the given 8 src pixels ? times upward
3321  */
3322 static inline void RENAME(duplicate)(uint8_t src[], int stride)
3323 {
3324 #ifdef HAVE_MMX
3325         asm volatile(
3326                 "movq (%0), %%mm0               \n\t"
3327                 "add %1, %0                     \n\t"
3328                 "movq %%mm0, (%0)               \n\t"
3329                 "movq %%mm0, (%0, %1)           \n\t"
3330                 "movq %%mm0, (%0, %1, 2)        \n\t"
3331                 : "+r" (src)
3332                 : "r" ((long)-stride)
3333         );
3334 #else
3335         int i;
3336         uint8_t *p=src;
3337         for(i=0; i<3; i++)
3338         {
3339                 p-= stride;
3340                 memcpy(p, src, 8);
3341         }
3342 #endif
3343 }
3344
3345 /**
3346  * Filters array of bytes (Y or U or V values)
3347  */
3348 static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3349         QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
3350 {
3351         PPContext __attribute__((aligned(8))) c= *c2; //copy to stack for faster access
3352         int x,y;
3353 #ifdef COMPILE_TIME_MODE
3354         const int mode= COMPILE_TIME_MODE;
3355 #else
3356         const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3357 #endif
3358         int black=0, white=255; // blackest black and whitest white in the picture
3359         int QPCorrecture= 256*256;
3360
3361         int copyAhead;
3362 #ifdef HAVE_MMX
3363         int i;
3364 #endif
3365
3366         const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
3367         const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
3368
3369         //FIXME remove
3370         uint64_t * const yHistogram= c.yHistogram;
3371         uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
3372         uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
3373         //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
3374
3375 #ifdef HAVE_MMX
3376         for(i=0; i<57; i++){
3377                 int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
3378                 int threshold= offset*2 + 1;
3379                 c.mmxDcOffset[i]= 0x7F - offset;
3380                 c.mmxDcThreshold[i]= 0x7F - threshold;
3381                 c.mmxDcOffset[i]*= 0x0101010101010101LL;
3382                 c.mmxDcThreshold[i]*= 0x0101010101010101LL;
3383         }
3384 #endif
3385
3386         if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
3387         else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
3388                 || (mode & FFMPEG_DEINT_FILTER)
3389                 || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
3390         else if(   (mode & V_DEBLOCK)
3391                 || (mode & LINEAR_IPOL_DEINT_FILTER)
3392                 || (mode & MEDIAN_DEINT_FILTER)
3393                 || (mode & V_A_DEBLOCK)) copyAhead=13;
3394         else if(mode & V_X1_FILTER) copyAhead=11;
3395 //        else if(mode & V_RK1_FILTER) copyAhead=10;
3396         else if(mode & DERING) copyAhead=9;
3397         else copyAhead=8;
3398
3399         copyAhead-= 8;
3400
3401         if(!isColor)
3402         {
3403                 uint64_t sum= 0;
3404                 int i;
3405                 uint64_t maxClipped;
3406                 uint64_t clipped;
3407                 double scale;
3408
3409                 c.frameNum++;
3410                 // first frame is fscked so we ignore it
3411                 if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
3412
3413                 for(i=0; i<256; i++)
3414                 {
3415                         sum+= yHistogram[i];
3416                 }
3417
3418                 /* we allways get a completly black picture first */
3419                 maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
3420
3421                 clipped= sum;
3422                 for(black=255; black>0; black--)
3423                 {
3424                         if(clipped < maxClipped) break;
3425                         clipped-= yHistogram[black];
3426                 }
3427
3428                 clipped= sum;
3429                 for(white=0; white<256; white++)
3430                 {
3431                         if(clipped < maxClipped) break;
3432                         clipped-= yHistogram[white];
3433                 }
3434
3435                 scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
3436
3437 #ifdef HAVE_MMX2
3438                 c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
3439                 c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
3440 #else
3441                 c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
3442                 c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
3443 #endif
3444
3445                 c.packedYOffset|= c.packedYOffset<<32;
3446                 c.packedYOffset|= c.packedYOffset<<16;
3447
3448                 c.packedYScale|= c.packedYScale<<32;
3449                 c.packedYScale|= c.packedYScale<<16;
3450
3451                 if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);
3452                 else                        QPCorrecture= 256*256;
3453         }
3454         else
3455         {
3456                 c.packedYScale= 0x0100010001000100LL;
3457                 c.packedYOffset= 0;
3458                 QPCorrecture= 256*256;
3459         }
3460
3461         /* copy & deinterlace first row of blocks */
3462         y=-BLOCK_SIZE;
3463         {
3464                 uint8_t *srcBlock= &(src[y*srcStride]);
3465                 uint8_t *dstBlock= tempDst + dstStride;
3466
3467                 // From this point on it is guranteed that we can read and write 16 lines downward
3468                 // finish 1 block before the next otherwise we might have a problem
3469                 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3470                 for(x=0; x<width; x+=BLOCK_SIZE)
3471                 {
3472
3473 #ifdef HAVE_MMX2
3474 /*
3475                         prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3476                         prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3477                         prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3478                         prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3479 */
3480
3481                         asm(
3482                                 "mov %4, %%"REG_a"              \n\t"
3483                                 "shr $2, %%"REG_a"              \n\t"
3484                                 "and $6, %%"REG_a"              \n\t"
3485                                 "add %5, %%"REG_a"              \n\t"
3486                                 "mov %%"REG_a", %%"REG_d"       \n\t"
3487                                 "imul %1, %%"REG_a"             \n\t"
3488                                 "imul %3, %%"REG_d"             \n\t"
3489                                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
3490                                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
3491                                 "add %1, %%"REG_a"              \n\t"
3492                                 "add %3, %%"REG_d"              \n\t"
3493                                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
3494                                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
3495                         :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3496                         "g" ((long)x), "g" ((long)copyAhead)
3497                         : "%"REG_a, "%"REG_d
3498                         );
3499
3500 #elif defined(HAVE_3DNOW)
3501 //FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
3502 /*                        prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3503                         prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3504                         prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3505                         prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3506 */
3507 #endif
3508
3509                         RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
3510                                 srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3511
3512                         RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
3513
3514                         if(mode & LINEAR_IPOL_DEINT_FILTER)
3515                                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3516                         else if(mode & LINEAR_BLEND_DEINT_FILTER)
3517                                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3518                         else if(mode & MEDIAN_DEINT_FILTER)
3519                                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
3520                         else if(mode & CUBIC_IPOL_DEINT_FILTER)
3521                                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3522                         else if(mode & FFMPEG_DEINT_FILTER)
3523                                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3524                         else if(mode & LOWPASS5_DEINT_FILTER)
3525                                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3526 /*                        else if(mode & CUBIC_BLEND_DEINT_FILTER)
3527                                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3528 */
3529                         dstBlock+=8;
3530                         srcBlock+=8;
3531                 }
3532                 if(width==FFABS(dstStride))
3533                         linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
3534                 else
3535                 {
3536                         int i;
3537                         for(i=0; i<copyAhead; i++)
3538                         {
3539                                 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
3540                         }
3541                 }
3542         }
3543
3544         for(y=0; y<height; y+=BLOCK_SIZE)
3545         {
3546                 //1% speedup if these are here instead of the inner loop
3547                 uint8_t *srcBlock= &(src[y*srcStride]);
3548                 uint8_t *dstBlock= &(dst[y*dstStride]);
3549 #ifdef HAVE_MMX
3550                 uint8_t *tempBlock1= c.tempBlocks;
3551                 uint8_t *tempBlock2= c.tempBlocks + 8;
3552 #endif
3553                 int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
3554                 int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
3555                 int QP=0;
3556                 /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
3557                    if not than use a temporary buffer */
3558                 if(y+15 >= height)
3559                 {
3560                         int i;
3561                         /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
3562                            blockcopy to dst later */
3563                         linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
3564                                 FFMAX(height-y-copyAhead, 0), srcStride);
3565
3566                         /* duplicate last line of src to fill the void upto line (copyAhead+7) */
3567                         for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
3568                                 memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
3569
3570                         /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
3571                         linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
3572
3573                         /* duplicate last line of dst to fill the void upto line (copyAhead) */
3574                         for(i=height-y+1; i<=copyAhead; i++)
3575                                 memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
3576
3577                         dstBlock= tempDst + dstStride;
3578                         srcBlock= tempSrc;
3579                 }
3580
3581                 // From this point on it is guranteed that we can read and write 16 lines downward
3582                 // finish 1 block before the next otherwise we might have a problem
3583                 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3584                 for(x=0; x<width; x+=BLOCK_SIZE)
3585                 {
3586                         const int stride= dstStride;
3587 #ifdef HAVE_MMX
3588                         uint8_t *tmpXchg;
3589 #endif
3590                         if(isColor)
3591                         {
3592                                 QP= QPptr[x>>qpHShift];
3593                                 c.nonBQP= nonBQPptr[x>>qpHShift];
3594                         }
3595                         else
3596                         {
3597                                 QP= QPptr[x>>4];
3598                                 QP= (QP* QPCorrecture + 256*128)>>16;
3599                                 c.nonBQP= nonBQPptr[x>>4];
3600                                 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
3601                                 yHistogram[ srcBlock[srcStride*12 + 4] ]++;
3602                         }
3603                         c.QP= QP;
3604 #ifdef HAVE_MMX
3605                         asm volatile(
3606                                 "movd %1, %%mm7         \n\t"
3607                                 "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
3608                                 "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
3609                                 "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP
3610                                 "movq %%mm7, %0         \n\t"
3611                                 : "=m" (c.pQPb)
3612                                 : "r" (QP)
3613                         );
3614 #endif
3615
3616
3617 #ifdef HAVE_MMX2
3618 /*
3619                         prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3620                         prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3621                         prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3622                         prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3623 */
3624
3625                         asm(
3626                                 "mov %4, %%"REG_a"              \n\t"
3627                                 "shr $2, %%"REG_a"              \n\t"
3628                                 "and $6, %%"REG_a"              \n\t"
3629                                 "add %5, %%"REG_a"              \n\t"
3630                                 "mov %%"REG_a", %%"REG_d"       \n\t"
3631                                 "imul %1, %%"REG_a"             \n\t"
3632                                 "imul %3, %%"REG_d"             \n\t"
3633                                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
3634                                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
3635                                 "add %1, %%"REG_a"              \n\t"
3636                                 "add %3, %%"REG_d"              \n\t"
3637                                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
3638                                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
3639                         :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3640                          "g" ((long)x), "g" ((long)copyAhead)
3641                         : "%"REG_a, "%"REG_d
3642                         );
3643
3644 #elif defined(HAVE_3DNOW)
3645 //FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
3646 /*                        prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3647                         prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3648                         prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3649                         prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3650 */
3651 #endif
3652
3653                         RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3654                                 srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3655
3656                         if(mode & LINEAR_IPOL_DEINT_FILTER)
3657                                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3658                         else if(mode & LINEAR_BLEND_DEINT_FILTER)
3659                                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3660                         else if(mode & MEDIAN_DEINT_FILTER)
3661                                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
3662                         else if(mode & CUBIC_IPOL_DEINT_FILTER)
3663                                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3664                         else if(mode & FFMPEG_DEINT_FILTER)
3665                                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3666                         else if(mode & LOWPASS5_DEINT_FILTER)
3667                                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3668 /*                        else if(mode & CUBIC_BLEND_DEINT_FILTER)
3669                                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3670 */
3671
3672                         /* only deblock if we have 2 blocks */
3673                         if(y + 8 < height)
3674                         {
3675                                 if(mode & V_X1_FILTER)
3676                                         RENAME(vertX1Filter)(dstBlock, stride, &c);
3677                                 else if(mode & V_DEBLOCK)
3678                                 {
3679                                         const int t= RENAME(vertClassify)(dstBlock, stride, &c);
3680
3681                                         if(t==1)
3682                                                 RENAME(doVertLowPass)(dstBlock, stride, &c);
3683                                         else if(t==2)
3684                                                 RENAME(doVertDefFilter)(dstBlock, stride, &c);
3685                                 }else if(mode & V_A_DEBLOCK){
3686                                         RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
3687                                 }
3688                         }
3689
3690 #ifdef HAVE_MMX
3691                         RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3692 #endif
3693                         /* check if we have a previous block to deblock it with dstBlock */
3694                         if(x - 8 >= 0)
3695                         {
3696 #ifdef HAVE_MMX
3697                                 if(mode & H_X1_FILTER)
3698                                         RENAME(vertX1Filter)(tempBlock1, 16, &c);
3699                                 else if(mode & H_DEBLOCK)
3700                                 {
3701 //START_TIMER
3702                                         const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3703 //STOP_TIMER("dc & minmax")
3704                                         if(t==1)
3705                                                 RENAME(doVertLowPass)(tempBlock1, 16, &c);
3706                                         else if(t==2)
3707                                                 RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3708                                 }else if(mode & H_A_DEBLOCK){
3709                                         RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
3710                                 }
3711
3712                                 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3713
3714 #else
3715                                 if(mode & H_X1_FILTER)
3716                                         horizX1Filter(dstBlock-4, stride, QP);
3717                                 else if(mode & H_DEBLOCK)
3718                                 {
3719 #ifdef HAVE_ALTIVEC
3720                                         unsigned char __attribute__ ((aligned(16))) tempBlock[272];
3721                                         transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
3722
3723                                         const int t=vertClassify_altivec(tempBlock-48, 16, &c);
3724                                         if(t==1) {
3725                                                 doVertLowPass_altivec(tempBlock-48, 16, &c);
3726                                                 transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3727                                         }
3728                                         else if(t==2) {
3729                                                 doVertDefFilter_altivec(tempBlock-48, 16, &c);
3730                                                 transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3731                                         }
3732 #else
3733                                         const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
3734
3735                                         if(t==1)
3736                                                 RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
3737                                         else if(t==2)
3738                                                 RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
3739 #endif
3740                                 }else if(mode & H_A_DEBLOCK){
3741                                         RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
3742                                 }
3743 #endif //HAVE_MMX
3744                                 if(mode & DERING)
3745                                 {
3746                                 //FIXME filter first line
3747                                         if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3748                                 }
3749
3750                                 if(mode & TEMP_NOISE_FILTER)
3751                                 {
3752                                         RENAME(tempNoiseReducer)(dstBlock-8, stride,
3753                                                 c.tempBlured[isColor] + y*dstStride + x,
3754                                                 c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3755                                                 c.ppMode.maxTmpNoise);
3756                                 }
3757                         }
3758
3759                         dstBlock+=8;
3760                         srcBlock+=8;
3761
3762 #ifdef HAVE_MMX
3763                         tmpXchg= tempBlock1;
3764                         tempBlock1= tempBlock2;
3765                         tempBlock2 = tmpXchg;
3766 #endif
3767                 }
3768
3769                 if(mode & DERING)
3770                 {
3771                                 if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
3772                 }
3773
3774                 if((mode & TEMP_NOISE_FILTER))
3775                 {
3776                         RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3777                                 c.tempBlured[isColor] + y*dstStride + x,
3778                                 c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3779                                 c.ppMode.maxTmpNoise);
3780                 }
3781
3782                 /* did we use a tmp buffer for the last lines*/
3783                 if(y+15 >= height)
3784                 {
3785                         uint8_t *dstBlock= &(dst[y*dstStride]);
3786                         if(width==FFABS(dstStride))
3787                                 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
3788                         else
3789                         {
3790                                 int i;
3791                                 for(i=0; i<height-y; i++)
3792                                 {
3793                                         memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3794                                 }
3795                         }
3796                 }
3797 /*
3798                 for(x=0; x<width; x+=32)
3799                 {
3800                         volatile int i;
3801                         i+=        + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
3802                                 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
3803                                 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
3804 //                                + dstBlock[x +13*dstStride]
3805 //                                + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
3806                 }*/
3807         }
3808 #ifdef HAVE_3DNOW
3809         asm volatile("femms");
3810 #elif defined (HAVE_MMX)
3811         asm volatile("emms");
3812 #endif
3813
3814 #ifdef DEBUG_BRIGHTNESS
3815         if(!isColor)
3816         {
3817                 int max=1;
3818                 int i;
3819                 for(i=0; i<256; i++)
3820                         if(yHistogram[i] > max) max=yHistogram[i];
3821
3822                 for(i=1; i<256; i++)
3823                 {
3824                         int x;
3825                         int start=yHistogram[i-1]/(max/256+1);
3826                         int end=yHistogram[i]/(max/256+1);
3827                         int inc= end > start ? 1 : -1;
3828                         for(x=start; x!=end+inc; x+=inc)
3829                                 dst[ i*dstStride + x]+=128;
3830                 }
3831
3832                 for(i=0; i<100; i+=2)
3833                 {
3834                         dst[ (white)*dstStride + i]+=128;
3835                         dst[ (black)*dstStride + i]+=128;
3836                 }
3837
3838         }
3839 #endif
3840
3841         *c2= c; //copy local context back
3842
3843 }